Keywords: Advanced Vehicle Safety Systems, Off-line and Online Data Processing Techniques, Human Factors in Intelligent Transportation Systems
Abstract: With the wide spread of artificial intelligence (AI) technologies, many applications using AIs are increasingly deployed in many fields. Specially anomaly detection is one of the key applications of AI. Among several targets, detecting anomaly behavior of drivers or vehicles has been attracting due to the growing demand of safety driving. It is crucial to study and evaluate techniques for anomaly driving detection with AI technologies. The Online Sequential Extreme Learning Machine (OS-ELM) is a recently attracting neural network model that has high memory efficiency and can perform high- speed sequential learning with streaming data. Though OS- ELM is known to be effective for anomaly detection, it has not yet been verified for non-stationary time series data such as driving sensor data. In this paper, we study the effectiveness of OS-ELM based anomaly driving behavior detector using sensor data of vehicles and compared the performance of it with a Hidden Markov Model (HMM) based method. Since the existing driving behavior benchmark data is not enough for evaluating anomaly driving, we also create a new dataset with a powered wheelchair. Throughout the evaluation, we show that the OS-ELM based anomaly driving detector has the same or even better accuracy in anomaly driving detection with much faster sequential learning speed compared with the HMM-based detector.

Keywords: Advanced Vehicle Safety Systems, Automated Vehicle Operation, Motion Planning, Navigation, Simulation and Modeling
Abstract: Recently, deep neural networks trained with Imitation-Learning techniques have managed to successfully control autonomous cars in a variety of urban and highway environments. One of the main limitations of policies trained with imitation learning that has become apparent, however, is that they show poor performance when having to deal with extreme situations at test time -- like high-speed collision avoidance -- since there is not enough data available from such rare cases during training. In our work, we take the stance that training complex active safety systems for vehicles should be performed in simulation and the transfer of the learned driving policy to the real vehicle should be performed utilizing simulation-to-reality transfer techniques. To communicate this idea, we setup a high-speed collision avoidance scenario in simulation and train the safety system with Reinforcement Learning. We utilize Domain Randomization to enable simulation-to-reality transfer. Here, the policy is not trained on a single version of the setup but on several variations of the problem, each with different parameters. Our experiments show that the resulting policy is able to generalize much better to different values for the vehicle speed and distance from the obstacle compared to policies trained in the non-randomized version of the setup.

Keywords: Advanced Vehicle Safety Systems, Other Theories, Applications, and Technologies
Abstract: Neural networks have become state-of-the-art for computer vision problems because of their ability to efficiently model complex functions from large amounts of data. While neural networks can be shown to perform well empirically for a variety of tasks, their performance is difficult to guarantee. Neural network verification tools have been developed that can certify robustness with respect to a given input image; however, for neural network systems used in closed-loop controllers, robustness with respect to individual images does not address multi-step properties of the neural network controller and its environment. Furthermore, neural network systems interacting in the physical world and using natural images are operating in a black-box environment, making formal verification intractable. This work combines the adaptive stress testing (AST) framework with neural network verification tools to search for the most likely sequence of image disturbances that cause the neural network controlled system to reach a failure. An autonomous aircraft taxi application is presented, and results show that the AST method finds failures with more likely image disturbances than baseline methods. Further analysis of AST results revealed an explainable cause of the failure, giving insight into the problematic scenarios that should be addressed.

Keywords: Advanced Vehicle Safety Systems, Driver Assistance Systems, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: Novel Advanced Driver Assistance Systems are being increasingly designed also for two-wheeled vehicles, both to increase safety and to improve the vehicle usability. One of the most challenging problems is ensuring roll stability at low-speed or standstill, useful to face dangerous conditions, for instance those related to impending rider’s sickness. In this work we develop an active gyroscopic controller to achieve stabilization and make the motorcycle sustain itself while at standstill. The parametric uncertainties that affect the system, the errors in the roll angle estimation, the unbalanced loads and the coupling between the roll dynamics and the rider’s posture make the control problem non trivial: thus a 3rd order sliding mode controller is designed and experimentally tested. Simulations allow to understand the tuning principle of the controller, while experimental results on an instrumented motorcycle equipped with a gyroscopic actuator prove that the proposed approach allows to achieve the desired vertical stabilization.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Electric Vehicles, Theory and Models for Optimization and Control
Abstract: This paper presents an automated vehicle speed planning system called the energy-optimal deceleration planning system (EDPS), which aims to maximize energy-recuperation of regenerative braking on connected and autonomous electrified vehicles. Based on the impending deceleration requirements resulting from speed reduction ahead of turning or stopping at a nearby intersection, a recuperation-energy-optimal speed profile is computed by maximizing the regenerative braking energy-efficiency while satisfying the physical limits of an electrified powertrain. In autonomous driving, the powertrain of an electrified vehicle can be directly controlled by the vehicle control unit (VCU) such that it follows the computed optimal speed profile. A practical force balance relationship based on an electrified powertrain is explicitly utilized to build the cost function of the associated optimal control problem. Optimal deceleration commands are determined by a parameterized polynomial-based deceleration model that is obtained by regression analyses with real-world test data from human drivers. To fast-track the validation of realistic EDPS performance, a virtual road environment, which includes connectivity infrastructure, is established, and the implementation EDPS results that are obtained by utilizing the connectivity information are presented and compared to human drivers' driving data. Compared with human drivers, we show that EDPS-based autonomous driving results in improved energy recuperation and a shorter driving time.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Theory and Models for Optimization and Control
Abstract: Stochastic Model Predictive Control has proved to be an efficient method to plan trajectories in uncertain environments, e.g., for autonomous vehicles. Chance constraints ensure that the probability of collision is bounded by a predefined risk parameter. However, considering chance constraints in an optimization problem can be challenging and computationally demanding. In this paper, we present a grid-based Stochastic Model Predictive Control approach. This approach allows to determine a simple deterministic reformulation of the chance constraints and reduces the computational effort, while considering the stochastic nature of the environment. Within the proposed method, we first divide the environment into a grid and, for each predicted step, assign each cell a probability value, which represents the probability that this cell will be occupied by surrounding vehicles. Then, the probabilistic grid is transformed into a binary grid of admissible and inadmissible cells by applying a threshold, representing a risk parameter. Only cells with an occupancy probability lower than the threshold are admissible for the controlled vehicle. Given the admissible cells, a convex hull is generated, which can then be used for trajectory planning. Simulations of an autonomous driving highway scenario show the benefits of the proposed grid-based Stochastic Model Predictive Control method.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Accurate Global Positioning, Sensing, Vision, and Perception
Abstract: This paper describes an openly available data set for rail vehicle positioning experiments. The data were collected using the DLR research vehicle RailDriVE on a segment of the harbor railway of Braunschweig, Germany, in February 2019. Several sensors of the RailDriVE equipment and an additional self-sufficient system provided by Technische Universität Darmstadt were employed, including two GNSS receivers, two inertial measurement units (IMU), and several speed and distance sensors (radar, optical, odometer) from the rail domain. Front-facing camera data has been included for documentation purposes. In order to simplify its use, some pre-processing steps were applied to the data, mainly to have common time and coordinate frames. Furthermore, example and reference positioning solutions have been included. The data set is described in detail, with information about the individual sensors and the data set structure (with parameters, raw, pre-processed, and reference data). Our work should be seen as a step towards more open and data-driven research in the rail domain, where experiments are difficult and costly. It is our hope to provide a solid basis for many different research efforts that provide the required technological advances for the rail sector.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation
Abstract: This paper addresses the challenge of safe path planning for autonomous vehicles at intersections. Rapidly-exploring Random Tree (RRT) as an effective local motion planning methodology has the ability to determine a feasible path. As the number of sampled positions increases, the probability of finding an optimal path increases. However, RRT is usually applied to the static environment due to its delay or lack of efficiency in planning a path to the goal area. In dynamic environments, redundant sampling positions near dynamic obstacles are not effective. Therefore, we proposed a methodology, pRRT, that combines Gaussian Processes Regression (GPR) and RRT to generate a local path to guide the vehicle through the intersection. The procedure includes two phases: prediction and planning. Under prediction, GPR predicts the vehicle's future trajectory points. The prediction results are combined with destination position (intersection exit) to generate a probability map for sampling such that position sample quality is increased by avoiding redundant samples. The optimal strategy is deployed to guarantee the trajectory is collision-free in both current and future time instances. A combination of both proposed improvements can thus result in a path that is collision-free under the dynamic intersection area. The proposed method also increased the speed of path generation compared to the RRT algorithm.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Travel Information, Travel Guidance, and Travel Demand Management, Simulation and Modeling
Abstract: Recently, efforts were made to standardize Signal Phase and Timing (SPaT) messages. Such messages contain the current signal phase with a prediction for the corresponding residual time for all approaches of a signalized intersection. Hence, the information can be utilized for the motion planning of human-driven/autonomously operated individual or public transport vehicles. Consequently, this leads to a more homogeneous traffic flow and a smoother speed profile. Unfortunately, adaptive signal control systems make it difficult to predict the SPaT information accurately. In this paper, we propose a novel machine learning approach to forecast the time series of residual times. A prediction framework that utilizes a Random Survival Forest (RSF) and a Long-Short-Term-Memory (LSTM) neural network is implemented. The machine learning models are compared to an Auto-Regressive Integrated Moving Average (ARIMA) and a Linear Regression (LR) model. For a proof of concept, the models are applied to a case study in the city of Zurich. Results show that the machine learning models outperform the traditional approaches, and in particular, the LSTM neural network is a promising tool for the enhancement of SPaT messages.

Keywords: Data Mining and Data Analysis, Incident Management, Other Theories, Applications, and Technologies
Abstract: There is a growing interest in understanding and identifying risky driving behaviours due to the numerous road fatalities attributed to them. For Heavy Goods Vehicles (HGVs), understanding driving behaviour and its impact on road safety is a subject of interest for researchers, the government and industrial sectors, as they rely on HGVs for the delivery of goods and services. The current literature on HGV driving behaviour uses machine learning techniques to uncover core driving incident stereotypes. However, human behaviour contains different levels of uncertainty and stereotyping driving behaviour with traditional crisp methods may cause information loss and establish unfair boundaries as they do not take context into consideration. Moreover, the sensor readings also have uncertainties, and the driving stereotypes may have different subjective interpretations. In order to capture those intermediate possibilities in driver stereotyping, we propose a data-driven Fuzzy Logic system that can capture the uncertainties within driving features (data) and between driving stereotypes, and classifies drivers according to the risk of their driving styles on a scale of 0 to 100, where 0 is a low risk driver and 100 high risk. The results from telematics data show that our proposed method provides a reliable, fair and explainable approach for real-time identification of HGV driving risk level.

Keywords: Data Mining and Data Analysis, Off-line and Online Data Processing Techniques, Simulation and Modeling
Abstract: In short-term traffic forecasting, the goal is to accurately predict future values of a traffic parameter of interest occurring shortly after the prediction is queried. The activity reported in this long-standing research field has been lately dominated by different Deep Learning approaches, yielding overly complex forecasting models that in general achieve accuracy gains of questionable practical utility. In this work we elaborate on the performance of Deep Echo State Networks for this particular task. The efficient learning algorithm and simpler parametric configuration of these alternative modeling approaches make them emerge as a competitive traffic forecasting method for real ITS applications deployed in devices and systems with stringently limited computational resources. An extensive comparison benchmark is designed with real traffic data captured over the city of Madrid (Spain), amounting to more than 130 Automatic Traffic Readers (ATRs) and several shallow learning, ensembles and Deep Learning models. Results from this comparison benchmark and the analysis of the statistical significance of the reported performance gaps are decisive: Deep Echo State Networks achieve more accurate traffic forecasts than the rest of considered modeling counterparts.

Keywords: Data Mining and Data Analysis, Off-line and Online Data Processing Techniques, Intelligent Logistics
Abstract: Scheduling and Routing in freight transport are usually the end products of an optimization process. However, the results may differ due to the heterogeneity of rules in different transport markets. Since the understanding of these decision rules is important for disaggregate freight modeling, this paper investigates development of an effective decision tree method for extracting them from an extensive freight transport data. We applied the method to model departure time and type of tours in freight transport of agricultural products. Having these two models together help us understand the whole anatomy of the freight activities for the selected transport segment. The models highlight the characteristics of time-of-day freight activities for this sector and indicate the importance of spatial and temporal characteristics in capturing the distinctions of the type of tours.

Keywords: Data Mining and Data Analysis
Abstract: Scenario-based methods for the assessment of Automated Vehicles (AVs) are widely supported by many players in the automotive field. Scenarios captured from real-world data can be used to define the scenarios for the assessment and to estimate their relevance. Therefore, different techniques are proposed for capturing scenarios from real-world data. In this paper, we propose a new method to capture scenarios from real-world data using a two-step approach. The first step consists in automatically labeling the data with tags. Second, we mine the scenarios, represented by a combination of tags, based on the labeled tags. One of the benefits of our approach is that the tags can be used to identify characteristics of a scenario that are shared among different type of scenarios. In this way, these characteristics need to be identified only once. Furthermore, the method is not specific for one type of scenario and, therefore, it can be applied to a large variety of scenarios. We provide two examples to illustrate the method. This paper is concluded with some promising future possibilities for our approach, such as automatic generation of scenarios for the assessment of automated vehicles.

Keywords: Data Mining and Data Analysis, Off-line and Online Data Processing Techniques, Data Management and Geographic Information Systems
Abstract: Many past efforts have been exerted towards describing and quantifying the effects of winter maintenance operations on traffic conditions and safety. As highly granular data on snowplow activity become available, many agencies are becoming interested in incorporating these data in their decision-making processes. However, due to its sheer volume, the processing of snowplow automatic vehicle location (AVL) data has been challenging. In addition, adverse weather conditions are usually accompanied by higher crash rates and also correlate with an increase in maintenance operations. Thus, improper model and variable selection can produce misleading results that indicate maintenance operations lead to a higher crash rate. This paper presents simple visualization tools and analysis methods that examine the effects of winter road maintenance operations on traffic safety by combining various data sources including weather, traffic, snowplow AVL, and crash data. Such intuitive tools and results can help agencies better understand the relationship between winter road maintenance activities and traffic safety.

Keywords: Driver Assistance Systems, Human Factors in Intelligent Transportation Systems, Advanced Vehicle Safety Systems
Abstract: Driver distraction is one of the main causes of traffic accidents, of which two critical types are cognitive distraction and visual distraction. To improve traffic safety, the functionality of detecting driver distraction is necessary for intelligent vehicles. However, while existing studies mainly applied classification-based methods, few efforts have been devoted on modelling the relationship between input features and time dependency of driver state, which is shown to be an effective way to improve accuracy. This study proposed a linear-chain conditional random fields (CRF) based approach to detect cognitive distraction and visual distraction. Experiment was carried out on a driving simulator to collect data, where n-back task and arrow task were used to induce cognitive and visual distraction, respectively. 4 types of interpretable features were applied, including mean of skin conductance level, standard deviation of horizontal gaze position, steering reversal rates and standard deviation of lateral position. The dynamic bayesian network (DBN) used in previous studies was introduced to be the baseline. Results showed that, the proposed CRF has a superior performance than DBN, with a holistic accuracy of 93.7% and average true positive rates of 91.2% and 89.2% for cognitive distraction and visual distraction, respectively. This performance gap is due to the incorporation of input features into the transition feature functions of the designed CRF, thus making it more suitable for modelling driver state transition pattern in real application.

Keywords: Driver Assistance Systems, Sensing, Vision, and Perception, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: Autonomous Vehicles (AVs) are currently being tested on both proving grounds and public roads. This includes the testing of partially, highly, and fully automated vehicles. Considering the level of complexity of these vehicles and their dependency on functionality of the on-board modules such as sensors, in conjunction with associated systems outside the vehicle including Vehicle-to-Everything (V2X), the importance of ensuring that each system works perfectly and independently and in collaboration with other systems is manifest. As the functionality of AVs depends on the performance of their sensors, this paper presents an overview of the AVs’ sensors and describes the role and function of each. In addition, different factors that impact the selection of a sensor for AVs are analyzed. Furthermore, the elements that should be considered for designing the test cases for AVs are suggested. This paper concludes by providing a list of proving grounds and a comparison of their facilities that have a special focus on AV testing and are capable of conducting essential tests with defined test elements.

Keywords: Driver Assistance Systems, Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems
Abstract: We consider the problem of intelligently navigating through complex traffic. Urban situations are defined by the underlying map structure and special regulatory objects of e.g. a stop line or crosswalk. Thereon dynamic vehicles (cars, bicycles, etc.) move forward, while trying to keep accident risks low. Especially at intersections, the combination and interaction of traffic elements is diverse and human drivers need to focus on specific elements which are critical for their behavior. To support the analysis, we present in this paper the so-called Risk Navigation System (RNS). RNS leverages a graph-based local dynamic map with Time-To-X indicators for extracting upcoming sharp curves, intersection zones and possible vehicle-to-object collision points. In real car recordings, recommended velocity profiles to avoid risks are visualized within a 2D environment. By focusing on communicating not only the positional but also the temporal relation, RNS potentially helps to enhance awareness and prediction capabilities of the user.

Keywords: Driver Assistance Systems, Cooperative Techniques and Systems, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: When the host car is equipped with predictive adaptive cruise control (ACC), the prediction of the preceding car's future speed is useful for improving the host car's performance, while getting the accurate prediction result is challenging as it requires a complex model with multiple information inputs. This paper presents a speed predictor based on a long short-term memory (LSTM) deep recurrent neural network (RNN), which utilizes various driving data (e.g., historical speed trajectories, the traffic light's statuses and road conditions) created by the traffic simulator VISSIM. The investigation results explain how the different information inputs affect the prediction's accuracy. Furthermore, the function of energy-optimal adaptive cruise control (EACC) based on model predictive control (MPC), which optimizes the host car's speed through taking the predicted speed of its preceding car into account, is presented in this work. The influence of predicted information on the performance of MPC-based EACC is also demonstrated in this work.

Keywords: Driver Assistance Systems, Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems
Abstract: Although intelligent Advanced Driving Assistance Systems and Autonomous Driving technologies are becoming ubiquituous, efficient and safe driving in dynamic, narrow or congested environment remains a theoretical and technological challenge. The reason mainly lies in the complexity of handling prediction and uncertainty, which are the fundamental ingredients for risk estimation. The theoretical shortcomings can be seen on the level of mesoscopic risk quantifiers like those based on Time-To-Event (TTX, e.g.~TTC, TTB) or spatial safe distances, which mainly reflect heuristic engineering practices or worst case deterministic forecasts of a scenario. In both cases, uncertainties are not explicitly considered and worst-case assumptions significantly reduce driving efficiency, like traffic throughput. However uncertainties are ubiquitous in real driving so that a fundamentally different strategy is required, which seeks for a tradeoff between risk and benefit. In this paper, we introduce a microscopic, probabilistic unified risk estimation framework ({bf PURE}), grounded in a first principles probabilistic risk theory, which generalizes to arbitrary risks and situations and which explicitly addresses these uncertainties. To underline the generality of the approach, we show how empirical risk measures like TTX can then be derived, understood and extended starting from the general risk estimation, making them interpretable in terms of their constituting parameters.

Keywords: Human Factors in Intelligent Transportation Systems
Abstract: Automated driving technology enhanced with automated mobility scooter is expected as a transport support for vulnerable elderly and disabled people. It is necessary to have a mobility strategy that can improve the safety, comfort, and acceptability of users for social implementation of automated mobility scooters. In this study, we specifically focus on the comfort of users. The hypotheses that the velocity of mobility scooters and the density of peripheral pedestrians will affect the comfort of users in automated driving and that it can be quantitatively evaluated by using their physiological responses were tested. The environment in which the mobility scooters and pedestrians move around together was reproduced using a driving simulator. The psychological and physiological responses of users during automated driving were investigated and recorded. The results showed that density of pedestrian affect the valence of users, and suggested that the possibility of quantitative evaluation of the user’s affect by using physiological measures.

Keywords: Human Factors in Intelligent Transportation Systems, Other Theories, Applications, and Technologies
Abstract: Trust determines public acceptance and uptake of autonomous vehicles (AV). Against popular assumption, trustin- automation is not a unitary construct, but comprises trust subtypes that have different behavioural properties and implications. Understanding trust subtypes—specifically competencebased trust (CT) and integrity-based trust (IT)—is crucial for improving public communication about AVs, analysing trustdependent driver behaviours and designing trust-recovering interfaces. However, these issues have been overlooked in most past research. As a pioneering step, the goal of this research was to analyse how experience with AV failures affect CT and IT. After experience with AV driving errors, both trust subtypes were reduced, with CT showing greater reduction. Structural equation modelling revealed CT to be the primary contributor to acceptance for driving automation, with stronger subsequent impact on preference for fully autonomous (SAE L5) than on semi-autonomous driving (SAE L3). These findings inform that trust-repairing interface should target CT after driving errors, especially for higher automation levels where humans are further removed from the loop. Future directions concerning CT– IT interactions, and the impact of AV anthropomorphic design and connected vehicle cyber-security on IT are discussed.

Keywords: Human Factors in Intelligent Transportation Systems, Driver Assistance Systems
Abstract: Trust and acceptance are key to the adoption of automated driving systems. However, recent research shows reluctance to trust and accept such systems. One way of overcoming the persisting barriers is to ensure good and appropriate user instruction. The present study investigates the influence of a text-based as opposed to a video-based instruction on trust and acceptance. To evaluate the influence of these instruction methods, the development of trust and acceptance were examined at different stages of system experience with an automated parking assistant. 36 participants took part in an on-road study, in which they performed a total of ten parking maneuvers. Trust and acceptance were assessed after the user instruction, but before the initial system experience, after five maneuvers and finally after five more subsequent maneuvers. Trust and acceptance ratings were higher in the video-based instruction group compared to the text-based introduction group before initial system experience. Furthermore, trust and acceptance increased with system usage. The assessments of the two groups converged over time and did not differ statistically after ten parking maneuvers. Nevertheless, the video-based introduction shows promise with regard to reducing prior distrust and rejection of automated driving and should therefore be considered as an instruction method.

Keywords: Human Factors in Intelligent Transportation Systems, Aerial, Marine and Surface Intelligent Vehicles, Driver Assistance Systems
Abstract: This study investigates the influence of different variations of active roll stabilization (curve tilting, roll compensation and passive) and rear wheel steering (with and without) on the severity of motion sickness (MS) symptoms experienced by vehicle occupants, as there is a hint for reduction of MS based on past studies. A simulator study with 50 participants (age 20-60, 50% women, 50% men) was conducted. The simulator used features eight DOF and is able to reproduce the maneuvers realistically, which is necessary to reduce the probability of cross couplings with simulator sickness. There were two scenarios: driving on a winding country road and highway driving with frequent lane changes. MS severity was measured subjectively via questionnaire and objectively via heartrate. In addition temperature data was collected. Even though MS symptoms have been evoked, no significant effects were found in case of suspension settings for MS in general. For single symptoms nausea and vertigo, slight significances were detected between the roll compensated car and the other variants. High lateral acceleration led to significantly higher motion sickness ratings for all suspension variants and should be avoided in trajectory planning for autonomous vehicles. In conclusion there is no benefit by the use of active roll stabilization and rear wheel steering systems in case of motion sickness, when driving in similar driving conditions as in this study. But there is also no negative effect and there is some evidence, that another kind of control algorithm may have a positive effect.

Keywords: Human Factors in Intelligent Transportation Systems
Abstract: Driver take-overs at system limits and the corresponding requests to intervene are not obligatory in highly automated driving. Therefore, minimal risk maneuvers may occur. In order to clearly communicate the automation status and the driver's task at such system limits, ambient light displays seem to have a high potential. Therefore, two ambient light displays were investigated in a driving simulation experiment, mounted either at the bottom of the windshield or on the steering wheel. Forty participants experienced two request to intervene scenarios, and two minimal risk maneuver scenarios during a highly automated drive. In general, both ambient light displays seem to ensure correct driver’s reactions and safety, and lead to positive subjective ratings. Results revealed no significant differences between the two ambient light display positions regarding the take-over behavior, gaze behavior and subjective rating, except in terms of the perceived brightness.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: Connected and Autonomous Vehicles (CAVs) are expected to bring major transformations to transport efficiency and safety. Studies show a range of possible impacts, from worse efficiency of CAVs at low penetration rates, to significant improvements in both efficiency and safety at high penetration rates and loads. However, existing studies tend to explore efficiency and safety separately, focus on one type of a road network, and include only cars rather than other vehicle types. This paper presents a comprehensive study on impact of CAVs on both efficiency and safety, in three types of networks (urban, national, motorway), simulating different penetration rates of vehicles with multiple levels of automation, using historical traffic data captured on Irish roads. Our study confirms existing results that near-maximum efficiency improvements are observed at relatively low penetration rates, but reveals further insights that the exact penetration ranges between 20% and 40% depending on the network type and traffic conditions. Safety results show a 30% increase of conflicts at lower penetration rates, but 50-80% reduction at higher ones, with consistent improvement for increased penetration. We further show that congestion has a higher impact on conflicts than penetration rates, highlighting the importance of unified evaluation of efficiency and safety.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Cooperative Techniques and Systems, Driver Assistance Systems
Abstract: Trajectory prediction is a crucial task required for autonomous driving. The highly interactions and uncertainties in real-world traffic scenarios make it a challenge to generate trajectories that are accurate, reasonable and covering diverse modality as much as possible. This paper propose a conditional Wasserstein auto-encoder trajectory prediction model (TrajCWAE) that combines the representation learning and variational inference to generate predictions with multi-modal nature. TrajCWAE model leverages a context embedder to learn the intentions among vehicles and imposes Gaussian mixture model to reconstruct the prior and posterior distributions. Wasserstein generative adversarial framework is then used to match the aggregated posterior distribution with prior distribution. Furthermore, kinematic constraints are considered to make the prediction physically feasible and socially acceptable. Experiments on two scenarios demonstrate that the proposed model outperforms state-of-the-art methods, achieving better accuracy, diversity and coverage.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Advanced Vehicle Safety Systems
Abstract: The widescale deployment of Autonomous Vehicles (AV) seems to be imminent despite many safety challenges that are yet to be resolved. It is well known that there are no universally agreed Verification and Validation (VV) methodologies to guarantee absolute safety, which is crucial for the acceptance of this technology. Existing standards focus on deterministic processes where the validation requires only a set of test cases that cover the requirements. Modern autonomous vehicles will undoubtedly include machine learning and probabilistic techniques that require a much more comprehensive testing regime due to the non-deterministic nature of the operating design domain. A rigourous statistical validation process is an essential component required to address this challenge. Most research in this area focuses on evaluating system performance in large scale real-world data gathering exercises (number of miles travelled), or randomised test scenarios in simulation.

This paper presents a new approach to compute the statistical characteristics of a system’s behaviour by biasing automatically generated test cases towards the worst case scenarios, identifying potential unsafe edge cases. We use reinforcement learning (RL) to learn the behaviours of simulated actors that cause unsafe behaviour measured by the well established RSS safety metric. We demonstrate that by using the method we can more efficiently validate a system using a smaller number of test cases by focusing the simulation towards the worst case scenario, generating edge cases that correspond to unsafe situations.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Human Factors in Intelligent Transportation Systems, Other Theories, Applications, and Technologies
Abstract: Safety is a decisive factor during the development of automotive systems. Modern vehicles are becoming more software-intensive, electronic components are increasingly replacing mechanical units. This is accompanied by a further increase in the complexity of the systems. Mobility concepts could be subject to fundamental changes in the future. There is a broad consensus among safety and security experts that traditional methods alone can no longer guarantee adequate safeguarding of software-intensive systems. Faced with the problems that the fundamental changes in today's designed systems require a need for new hazard analyses, Leveson developed the System Theoretic Process Analysis (STPA) in 2004. This paper shows how the STPA analysis can be used as a valuable tool to identify potential hazards. In this paper partial results of STPA for a vehicle SAE level five are presented and compared with the results of STPA, carried out for a vehicle SAE level four by the same authors. This paper has not yet been published but a draft version is available from the authors.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Theory and Models for Optimization and Control, ITS Field Tests and Implementation
Abstract: This paper presents a multi-scenario full-range speed lateral automated vehicle controller. A speed-dependent LPV model is designed to deal with two different situations: 1)vehicle tracking capabilities to follow a pre-defined trajectory; and 2) vehicle response to sudden reference changes as occur either when activating the automated system for the first time or when performing a lane-change. The proposed solution is based on the Linear Parameter Varying (LPV) control approach, where an output-feedback dynamical controller is designed based on the Linear Matrix Inequalities (LMIs). The control synthesis is carried out using the Linear Fractional Transformation approach, to reduce the conservatism, combined with the H∞ control problem. Simulation results show the tracking performance and the smoothness of the control inputs which provides a comfortable riding. Finally, the algorithm has been implemented on a robotized Renault ZOE and validates on test tracks, providing encouraging results.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Driver Assistance Systems, Simulation and Modeling
Abstract: The development and validation of automated driving functions requires the simulation of numerous traffic situations. The introduction of scenario-based development and testing approaches emphasises the importance of simulation scenarios. For an effective evaluation of simulation scenarios, a criticality evaluation with respect to the behaviour of the automated driving function is necessary. Therefore, this work presents a multidimensional criticality evaluation for the eval- uation of traffic situations. The methodology summarises well- known criticality metrics and assesses the overall criticality of the situation. First, individual metrics are calculated for the ego- vehicle. Subsequently, potential traffic conflicts are evaluated and summarized with appropriate relational criticality metrics. Lastly, the criticalities of further objects are assessed with the identical method and included in the overall criticality. A proof of concept was implemented in accordance with the methodology. The results are presented and discussed upon scenario-based approaches towards automated driving function development and testing. Finally, the work is concluded and future work is pointed out.

Keywords: Sensing, Vision, and Perception
Abstract: The fusion of sensor data from heterogeneous sensors is crucial for robust perception in various robotics applications that involve moving platforms, for instance, autonomous vehicle navigation. In particular, combining camera and lidar sensors enables the projection of precise range information of the surrounding environment onto visual images. It also makes it possible to label each lidar point with visual segmentation/classification results for 3D mapping, which facilitates a higher level understanding of the scene. The task is however considered non-trivial due to intrinsic and extrinsic sensor calibration, and the distortion of lidar points resulting from the ego-motion of the platform. Despite the existence of many lidar ego-motion correction methods, the errors in the correction process due to uncertainty in ego-motion estimation are not possible to remove completely. It is thus essential to consider the problem a probabilistic process where the ego-motion estimation uncertainty is modelled and considered consistently. The paper investigates the probabilistic lidar ego-motion correction and lidar-to-camera projection, where both the uncertainty in the ego-motion estimation and time jitter in sensory measurements are incorporated. The proposed approach is validated both in simulation and using real-world data collected from an electric vehicle retrofitted with wide-angle cameras and a 16-beam scanning lidar.

Keywords: Sensing, Vision, and Perception
Abstract: Semantic segmentation of images enables pixel-wise scene understanding which in turn is a critical component for tasks such as autonomous driving. While recent implementations of semantic image segmentation have achieved remarkable accuracy, misclassifications remain inevitable. For safety-critical tasks such as free-space computing, it is desirable to know when and where the segmentation will fail. We propose using the concept of introspection to predict the failures of a given semantic segmentation model. A separate introspective model is trained to predict the errors of a given model. This is accomplished by training the given model with the errors made on a set of previous inputs. By using the same architecture for the introspective model as for the semantic segmentation, the proposed model learns to predict pixel-wise failure probabilities. This allows to predict both when and where the semantic segmentation will fail. Sharing the feature encoder with the inspected model reduces training and inference time while increasing performance. We evaluate our approach on the large-scale A2D2 driving data set. In a precision-recall analysis, the proposed method outperforms two state-of-the-art uncertainty estimation methods by 3.2% and 6.7% while requiring significantly less resources during inference. Additionally, combining introspection with a state-of-the-art method further increases the performance by up to 3.7%.

Keywords: Sensing, Vision, and Perception, Sensing and Intervening, Detectors and Actuators, Data Mining and Data Analysis
Abstract: Identifying interactions of vehicles on the road is important for accident analysis and driving behavior assessment. Our interactions include those with passing/passed, cut-in, crossing, frontal, on-coming, parallel driving vehicles, and ego-vehicle actions to change lane, stop, turn, and speeding. We use visual motion recorded in driving video taken by a dashboard camera to identify such interaction. Motion profiles from videos are filtered at critical positions, which reduces the complexity from object detection, depth sensing, target tracking, and motion estimation. The results are obtained efficiently, and the accuracy is also acceptable. The results can be used in driving video mining, traffic analysis, driver behavior understanding, etc.

Keywords: Sensing, Vision, and Perception, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Human Factors in Intelligent Transportation Systems
Abstract: Today’s vehicles are capable of detecting environmental traffic participants, such as other vehicles, pedestrians, traffic lights etc, and communicating with each other or infrastructures. Typical on-board detectors include LiDAR, camera and so on. These vehicles which can make driving decisions based on the detected information without human intervention are named CAV (connected and autonomous vehicles). However, in a long period, the road traffic is mixed by traditional vehicles (human driven vehicles, or HVs) and CAV. The system can only “see” the near field vehicles around the CAVs by means of on-board detectors or VANET (vehicular ad hoc network). Far-field vehicles are either too far away or covered by near-field vehicles. In order to enhance the sensing capabilities of VANET or CAV, the manuscript propose a far-field vehicles sensing method, called F2-sensing. The method combines the deep learning and the car following logic. The rationale is that, as the vehicles react to downstream vehicles’ states variation, when the CAVs and the near field vehicles’ states are known, the downstream vehicles’ existence and its real-time location can be estimated. The proposed method is tested against real world dataset, which proves the usefulness of the method.

Keywords: Sensing, Vision, and Perception, Driver Assistance Systems
Abstract: Computer vision tasks such as semantic segmentation perform very well in good weather conditions, but if the weather turns bad, they have problems to achieve this performance in these conditions. One possibility to obtain more robust and reliable results in adverse weather conditions is to use video-segmentation approaches instead of commonly used single-image segmentation methods. Video-segmentation approaches capture temporal information of the previous video-frames in addition to current image information, and hence, they are more robust against disturbances, especially if they occur in only a few frames of the video-sequence. However, video-segmentation approaches, which are often based on recurrent neural networks, cannot be applied in real-time applications anymore, since their recurrent structures in the network are computational expensive. For instance, the inference time of the LSTM-ICNet, in which recurrent units are placed at proper positions in the single-segmentation approach ICNet, increases up to 61 percent compared to the basic ICNet. Hence, in this work, the LSTM-ICNet is sped up by modifying the recurrent units of the network so that it becomes real-time capable again. Experiments on different datasets and various weather conditions show that the inference time can be decreased by about 23 percent by these modifications, while they achieve similar performance than the LSTM-ICNet and outperform the single-segmentation approach enormously in adverse weather conditions.

Keywords: Simulation and Modeling, Data Mining and Data Analysis, Theory and Models for Optimization and Control
Abstract: Due to the well-known environmental issues, European CO2 emission standards are more and more stringent. However a direct measurement of these CO2 emissions is actually unachievable. Our work aims to predict such a real-world fuel consumption by modeling its three influencing factors: the vehicle, the driving behavior and the use case. The vehicle factor is modeled through a quasi-static model, parametrized by vehicle parameters only (i.e. mass body, engine power, etc.). Driving behavior and use case are modeled thanks to a data analysis of a real-world vehicles and trips database by performing a classification on features derived from speed and acceleration trace recordings. Our approach was evaluated on trips of 162 vehicles and compared to real-fuel consumption gathered in an online self-report fuel tracking service database, by evaluating the gap between the official fuel consumption and the real one given by the database or our model, according to the identified driving behaviors and use cases. A maximal difference of around 10% is recovered for 85% of the vehicles and 20% difference for 96% of the vehicles.

Keywords: Simulation and Modeling, ITS Policy, Design, Architecture and Standards, Ride Matching and Reservation
Abstract: This paper studies the impacts of driver classification regulations on the profitability of transportation network companies (TNCs). These impacts are assessed by a market-equilibrium model that captures the arrival of passengers and drivers and quantifies the TNC's pricing decisions in the presence/absence of regulation. We analyze how TNC profit is affected by the driver classification regulations due to two effects: (a) increase in driver wages due to minimum wage and overtime compensation requirement; (b) loss of flexibility of driver working schedules. We deconvolved the contribution of these two effects on TNC profit reduction and estimated each of these effects using empirical data from San Francisco. Our results suggest that the increased driver wage effect contributes substantially to the loss of TNC profit (45% of original profit). On the other hand, the loss of flexibility effect varies significantly (6-41% of original profit) depending on the flexibility of driver working schedules. We also evaluated the impact of the proposed regulation on TNC market outcomes such as drivers earnings, utilization rate, trip fares and total number of trips.

Keywords: Simulation and Modeling
Abstract: Simulation-based validation methods exhibit properties which should allow them to contribute to the release of highly automated driving functions (HADF). Monte-Carlo (MC) simulations make it possible to generate statistical statements about a HADF. However, the fact that accidents happen very rarely in the real world can strongly enlarge the number of MC simulation runs which are needed to obtain statistically stable results. In the past, importance sampling (IS) was used to reduce these simulation costs. This paper considers some of the still existent limitations of IS: The definition of proper safety metrics needed to find good IS distributions and the application of IS to high dimensional parameter spaces. The approach in this paper was able to reduce an IS optimization of 2700 dimensions to 17 optimizations of 9 dimensions. The number of MC runs could be reduced by a factor of 350.

Keywords: Simulation and Modeling, Travel Information, Travel Guidance, and Travel Demand Management, Network Modeling
Abstract: This paper studies the problem of estimation and computation of reliable least-time paths in stochastic time-varying (STV) networks with spatio-temporal dependencies. For a given desired confidence level α, the least-time paths from any origin to a given destination node are to be found over a desired planning horizon. In STV networks, least-time path finding approaches aim to incorporate an element of reliability to help travelers better plan their trips to prepare for the risk of arriving later or traveling for longer than desired. A label-correcting algorithm that incorporates time-dependence of the travel time distributions is proposed. The algorithm incorporates a Monte Carlo sampling approach for a path travel time estimation with time-dependence, which can also be used as an approximate solution method with spatial link travel-time correlations. Numerical results on the large-scale Chicago network are provided to test for the performance of the algorithms and the robustness of solutions. The trade-off between accuracy and efficiency of the approximate solution method compared to a Monte Carlo simulation-based approach is discussed and evaluated.

Keywords: Simulation and Modeling, Public Transportation Management, Rail Traffic Management
Abstract: If not properly managed neither planned on real needs, the maintenance of rolling stock may strongly affect rail operations in local public transport, risking to compromise the quality of service or generating an over sizing of the fleet. Therefore, an effective coordination is required between the Operation and Maintenance departments. Some flexibility in maintenance activities – i.e., preventive and on condition maintenance policies - has already been applied for some years in the regional rail transport with successful results; however, it has not been introduced yet in rail public transport, where a corrective maintenance is generally adopted. In this work, the proper scheduling of more flexible maintenance activities in the rail public transport context is addressed through the use of discrete event simulation. Real data sets provided by the Italian GTT-Gruppo Torinese Trasporti company have been used to test the proposed approach and to carry out a multi-scenarios campaign, aiming at analyzing the effectiveness of the maintenance process when certain operating conditions change or unexpected events occur. Some improvement proposals have also been analyzed with the proposed simulation method.

Keywords: Theory and Models for Optimization and Control, Traffic Theory for ITS
Abstract: The paper provides conditions that guarantee existence and uniqueness of classical solutions for a non-local conservation law on a ring-road with nudging (or “look behind”) terms. The obtained conditions are novel, as they are not covered by existing results in the literature. The paper also provides results which indicate that nudging can increase the flow in a ring-road and, if properly designed, can have a strong stabilizing effect on traffic flow. The efficiency of the use of nudging terms is demonstrated by means of a numerical example.

Keywords: Theory and Models for Optimization and Control, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: Reinforcement learning (RL) has achieved remarkable performance in numerous sequential decision making and control tasks. However, a common problem is that learned nearly optimal policy always overfits to the training environment and may not be extended to situations never encountered during training. For practical applications, the randomness of environment usually leads to some devastating events, which should be the focus of safety-critical systems such as autonomous driving. In this paper, we introduce the minimax formulation and distributional framework to improve the generalization ability of RL algorithms and develop the Minimax Distributional Soft Actor-Critic (Minimax DSAC) algorithm. Minimax formulation aims to seek optimal policy considering the most severe variations from environment, in which the protagonist policy maximizes action-value function while the adversary policy tries to minimize it. Distributional framework aims to learn a state-action return distribution, from which we can model the risk of different returns explicitly, thereby formulating a risk-averse protagonist policy and a risk-seeking adversarial policy. We implement our method on the decision-making tasks of autonomous vehicles at intersections and test the trained policy in distinct environments. Results demonstrate that our method can greatly improve the generalization ability of the protagonist agent to different environmental variations.

Keywords: Theory and Models for Optimization and Control
Abstract: Shippers usually procure transportation services from carriers via combinatorial auctions to reduce costs and improve service levels. Compared with single-round sealed bid auctions, multi-round auctions can reduce the revelation of confidential cost information of the carriers. In this paper, a two-phase multi-round combinatorial auction mechanism is proposed for truckload transportation service procurement, in which each transportation request is represented by a lane. The first phase is a combinatorial clock auction, which is terminated when the prices of the lanes are raised high enough such that each lane is bid by at least one carrier. In the second phase, the auctioneer provides some supplementary bundles of requests open for bid and adjusts the prices of the bundles, and each carrier decides whether to bid for some of the bundles in addition to its bids submitted in the first phase. Computational results show that the proposed mechanism can achieve an optimal or a near-optimal allocation of the requests to the carriers in terms of social efficiency.

Keywords: Theory and Models for Optimization and Control, Automated Vehicle Operation, Motion Planning, Navigation, Data Mining and Data Analysis
Abstract: The system architecture of an autonomous vehicle consists of several parts. One of those is motion control, responsible for following a desired trajectory or path. To fulfill the task of trajectory following using a future trajectory reference while considering constraints, model predictive control (MPC) is especially suitable. MPC in general is however computationally heavy and the control performance mainly relies on the accuracy of the model used. This is why it is important to find an accurate model of the considered dynamics which is still relatively easy to evaluate online. In this paper we present a nonlinear MPC approach based on neural network models. Our aim is to control the longitudinal position of an autonomous passenger vehicle with a complex conventional drivetrain. Different network outputs and topologies are discussed and compared in terms of their ability to accurately predict the longitudinal position of the vehicle. The modeling approaches investigated are three different three layer perceptron (TLP) models. Among those, a multi model neural network approach provides the best results and is therefore used as a prediction model within a nonlinear MPC. Simulation results using a complex high fidelity vehicle model as a plant are given and discussed.

Keywords: Theory and Models for Optimization and Control, Traffic Theory for ITS, Other Theories, Applications, and Technologies
Abstract: In order to control and reduce the congestion in motorway traffic networks, it is required to measure important traffic variables, e.g. densities, that can be observed by sensors. However, to reduce the operational costs, one should look for the most efficient methods to place the minimum number of sensors in a given network. In this paper, we discuss the structural observability of a traffic system, namely, the density dynamics defined on a motorway ring road. For this purpose, LWR theory in a spatial discretization form is employed, and the nonlinear dynamics of traffic density associated with different cells of the network have been derived. Then, by considering a linearization of the ordinary differential equations (ODEs), we derive the minimum number of sensors that are needed to render the network weakly or strongly structurally observable. In this framework, the parameters of the system can have any nonzero value, and the exact value of nonzero elements (weights) is not of interest. In this work, we also discuss optimal locations in the traffic network to place the minimum set of sensors.

Keywords: Theory and Models for Optimization and Control
Abstract: Due to unexpected events such as bad weather, unknown object intrusion, train malfunction, etc., the high-speed train arrival or departure time is often delayed, which makes the train unable to operate according to the initial train timetable. At present, the dynamic scheduling problem of high-speed trains under arrival/departure time delay is lack of in-depth analysis, which results in excessive adjustment of the initial train timetable by existing dynamic scheduling methods. This paper aims to establish a delay propagation mechanism model for high-speed train operation based on the Reachable Matrix for the train scheduling problem. Buffer Matrix and the Conflict Matrix are used to analyze the influence range and influence degree on the initial train timetable. The actual operation data of Nanjing South to Cangzhou section is selected to verify the proposed model. The simulation results show that the model can analyze the affected train range and influence degree, which provides scientific basis for the adjustment of high-speed train timetable, and ensure the scheduling plan continuity and stability.

Keywords: Rail Traffic Management, Public Transportation Management, Simulation and Modeling
Abstract: Real-time railway traffic management is important for the daily operations of railway systems. It predicts and resolves operational conflicts caused by events like excessive passenger boardings/alightings. Traditional optimization methods for this problem are restricted by the size of the problem instances. Therefore, this paper proposes a reinforcement learning-based timetable rescheduling method. Our method learns how to reschedule a timetable off-line and then can be applied online to make an optimal dispatching decision immediately by sensing the current state of the railway environment. Experiments show that the rescheduling solution obtained by the proposed reinforcement learning method is affected by the state representation of the railway environment. The proposed method was tested to a part of the Dutch railways considering scenarios with single initial train delays and multiple initial train delays. In both cases, our method found high-quality rescheduling solutions within limited training episodes.

Keywords: Rail Traffic Management, Theory and Models for Optimization and Control, Simulation and Modeling
Abstract: This paper describes a new approach to develop a real-world automated scheduler applicable for Australian sugarcane industry. In Australia, the transport sector plays a critical role in raw sugarcane harvest and accounts for over 35% of the total cost of raw sugar production. The generation of an optimised schedule can bring the following practical benefits: eliminate bin supply delays to harvesters, minimise the number of locomotives/bins, reduce the locomotive shifts, control the sugarcane age/quality, etc. To generate such a scheduler, a new optimisation approach is developed based on job shop scheduling techniques using constraint programming and mixed integer programming. The proposed approach can produce solutions for small-scale and large-scale cases in agriculture/crops transport systems in a reasonable time. Mixed integer programming focuses on objective function using linear relaxation to prune suboptimal solutions, while constraint programming focuses on the model using filtering algorithms to eliminate infeasible candidate solutions. The applicability of the developed scheduler has been validated by a real-world case study for Kalamia Mill in Queensland, Australia. Following from the validation and discussion, it is concluded that the automated scheduler would be a valuable optimisation tool for transport modellers in Australian sugarcane industry.

Keywords: Data Mining and Data Analysis, Simulation and Modeling, Other Theories, Applications, and Technologies
Abstract: With the accelerated development of cities, the traffic capacity cannot catch up with traffic rising. The urban rail transit system is facing severe challenges. Accurate prediction of passenger flow can help optimize the operation plan and improve operation efficiency. Traditional machine learning-based intelligent control methods are restricted by insufficient data. Owing to lacking effective incentives and trust, data from different urban rail lines or operators cannot be shared directly. In this paper, we propose a distributed federal learning method for accurate prediction of rail transit passenger flow based on blockchain. The proposed method performs distributed machine learning without a trusted central server. The blockchain smart contract is used to realize the management of the entire federal learning. Considering the limitations of the traditional time series model, we choose the distributed long and short term memory (LSTM) networks as the supervised learning model for passenger flow prediction. In addition, we establish an incentive mechanism to reward those participants who contribute to the model. The simulation results demonstrate high efficiency and accuracy of our proposed intelligent control method.

Keywords: Commercial Fleet Management, Intelligent Logistics, Theory and Models for Optimization and Control
Abstract: Truck drivers often have difficulty finding appropriate rest locations due to truck parking shortages, which can significantly impact drivers' safety, industry costs, and the environment. Nevertheless, the lack of truck parking availability information makes it hard to account for parking during planning, leading research on truck routing and scheduling to usually assume that parking facilities are always available. In this paper, we propose a model that accounts for parking availability when planning long-haul truck shipments in addition to working hours constraints. The proposed method plans a minimum cost path and schedule such that the itinerary is regulation-compliant, and parking is guaranteed at all scheduled stops. The problem is modeled as a resource-constrained shortest path problem, and a label correcting algorithm is used to find a near-optimal solution. Computational experiments are used to compare the cost of solutions that use parking availability information with ones that do not by simulating drivers' behavior when parking is not available. Results show that, when parking constraints are imposed, parking availability levels can significantly affect costs. However, if irregular parking potential costs are factored in, the prevented penalty costs can exceed (or at least partially offset) the cost increase caused by parking constraints.

Keywords: Commercial Fleet Management, Roadside and On-board Safety Monitoring
Abstract: Questionnaire survey, such as Manchester Driver Behavior Questionnaire (DBQ) and in-vehicle trajectory data are both valid resources to identify risky drivers. Investigating the relationships between self-reported driving behavior and psychology and observed risky driving behavior from in-vehicle trajectory data can provide better understanding of personal factors contributing to risky driving, allowing the more effective development of safety education and road management countermeasures and interventions. This paper analyzed GPS trajectory obtained from 723 professional online car-hailing drivers. Through road type matching, risky driving feature extraction and clustering analysis, each driver was given a risky driving level. The level was then compared with their self-reported driving behavior, psychology statues such as anxiety and driving anger, as well as social factors including safety culture, occupation health, and number of complaints. Results show that social factors are more relative to driver’s risk driving level than other self-reported factors. This suggest improving safety culture of company and enhance occupation health of drivers will be effective to reduce risky driving of online car-hailing drivers.

Keywords: Commercial Fleet Management, Electric Vehicles, Infrastructure for Charging, Communication and Controls
Abstract: The electric vehicle market is booming. However, these vehicles need to be refilled more often and do so much more slowly than internal combustion engine (ICE) vehicles. The arrival of autonomous vehicles will enable both fully centralised systems for taxi fleet management and a 24/7 use of each taxi. Finally, the ride-sharing market is also booming. Thus, efficient future taxi fleets will have to provide efficient, integrated solutions for ride-sharing, charging and automation. In this paper, the problem focused on is a variation of the Dial-A-Ride-Problem (DARP) where charging as well as the availability of charging stations are taken into account: Given a fleet of autonomous and electric taxis, a charging infrastructure, and a set of trip requests, the objective is to assign trips and charges to taxis such that the total profit of the fleet is maximised. Our contribution consists in the development of a greedy method, and of a simulated annealing. Our methods are evaluated on large instances (10000 requests) based on taxi trip datasets in Porto. Our conclusions show that while high-capacity batteries are largely unneeded in normal circumstances, they are capital in case of disruption, and useful when the charging infrastructure is shared, with queueing time to access to a charger. Parking searches also represent a significant energy expense for autonomous taxis.

Keywords: Commercial Fleet Management, Theory and Models for Optimization and Control, Public Transportation Management
Abstract: In this paper, we present a data-driven stochastic programming model for reducing car-sharing relocation cost under uncertain customer demands. Instead of using parametric methods to estimate demand probability distributions, we propose an integration of non-parametric kernel density estimation, sample average approximation and a two-stage stochastic programming model. The proposed approach computes high-quality car-sharing relocation solutions by better leveraging the information provided by large-scale historical data. To validate the performance of the proposed approach, we conduct numerical experiments using the New York taxi trip data sets. Our results show that the proposed approach outperforms the parametric approach using Laplace and Poisson distributions and the deterministic model in terms of profit and combined holding and relocation costs. Most importantly, it reduces on average more than 50% of relocation rate compared with the parametric method and 67% of relocation rate compared with the deterministic model.

Keywords: ITS Field Tests and Implementation, Commercial Fleet Management, Intelligent Logistics
Abstract: Automated solutions for executing services in urban areas have become a trend in the past years. Examples of such services are: package delivery, transportation, street cleaning, waste disposal or vegetation care. They are also part of new concepts of smart cities. The vision is to have a vehicle fleet that provides these services at demand of inhabitants or authorities in urban areas. These vehicles have the capability to drive autonomously and interact with other traffic participants in order to accomplish a specific task. An important aspect is the mission planning of the vehicles. We present a concept of an interactive planning and management of a vehicle fleet that executes requested service demands in urban areas and the interaction between the involved stakeholders. The service requester gets immediate response regarding their request and can track, change or cancel it with immediate adaption of the plan. The concept was implemented and the system was used a few months for delivery services during a real world laboratory in a new built district in Heilbronn (Germany), with about 800 inhabitants.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation
Abstract: The application of motion planning for autonomous vehicles has been primarily focused either in highly structured or unstructured environments. However, many environments in the real-world share the characteristics of both and can be classified as semi-structured. The adaptation of the strategies from other environments to that of semi-structured, although possible, do not produce trajectories with the required characteristics, especially when the environment is dynamic. To that end, this paper introduced a two-level hierarchical planning strategy consisting of a discrete lane-network-based global planner and a Hybrid A* local planner that: (i) generates a smooth, safe and kinematically feasible path in real-time; (ii) considers structural constraints of the environment from an a priori map. Furthermore, a valid lane-network-based sub-goaling strategy is proposed for providing a reference goal during the local planning process. Simulation and live tests have been conducted to evaluate the functionality of the strategy in several case studies. The implementation choices of: (i) using an open-source highly automated driving (HAD) map framework, Lanelet2, (ii) developing as plugins to an open-source navigation stack, Move Base Flex (MBF), allow the proposed strategy to be easily adopted on other autonomous platforms.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Theory and Models for Optimization and Control, Modeling, Simulation, and Control of Pedestrians and Cyclists
Abstract: One of the difficult tasks associated with driving in suburban or urban roads is the interactions with pedestrians. One often finds it hard to judge the chances of a pedestrian or bicycle suddenly turning onto the driving path. These leads to a natural slow down response by the drivers. Since these responses are based on the risk feeling of driver, they are probabilistic in nature. This study shares a scenario-based model predictive control algorithm considering probabilistic constraint (PSMPC) to handle such pedestrian interactions. An Interacting Multiple-Model Kalman Filter (IMM-KF) is used to predict the pedestrian path as multiple trajectories of independent probabilities. The task is formulated into a nonlinear MPC problem. We use a non-linear optimization solver named Interior Point OPTimizer(IPOPT). We introduce a modified form of inverse square root unit function to represent the collision probability into a deterministic function that is compatible with IPOPT. Having simulated it in MATLAB, the controller gives a very natural control behaviour for shared road driving compared to single scenario deterministic MPC.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Cooperative Techniques and Systems
Abstract: Automated vehicles allow to adapt new parking concepts, which have the potential to increase the parking capacity significantly. Thereby, the abilities to assign optimal parking spaces to the vehicles and to perform shunting operations are systematically utilised. In this paper, the typical parking order with spots of constant width is replaced by a loose parking order with variable width. In this way, the width of each individual vehicle can be considered, which leads to an improved utilisation of the available space. Despite this improvement, first simulation studies without shunting operations under realistic conditions did not show the expected results. Therefore, an optimum strategy for shunting operations is deployed. As the optimisation is computationally challenging, a heuristic method is proposed which shows similar results as the optimal solution. A substantial increase in the parking capacity of up to 12% for a commercial parking garage is achieved, where no constructional modifications are necessary.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems, Simulation and Modeling
Abstract: Highly automated driving systems need to master a highly complex environment and are required to show meaningful behavior in any situation occurring in mixed traffic with humans. Deriving a sufficiently complete and consistent set of system-level requirements capturing all possible traffic situations is a significant problem that has not been solved in existing literature. In this paper, we propose a new method called SOCA addressing this problem by introducing a novel abstraction of traffic situations, called zone graph, and using this abstraction in a morphological behavior analysis. The morphological behavior analysis enables us to derive a set of system-level requirements with guarantees on completeness and consistency. We illustrate our method on a slice-of-reality example from the automated driving domain.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, ITS Policy, Design, Architecture and Standards
Abstract: Apart from electrification, automated driving has become a large area of research in the automotive industry. In contrast to the mature and stable state of the art development in automotive industry, the development of new automated driving functions implies the integration of completely new features. This brings great uncertainties into concept development, bearing the risk of delay during integration. In our work we present a best practice example for the development of an automated driving function concept. We show how the integration risk can be reduced by defining additional requirements for the single software components of the automated driving function to account for their interaction. We demonstrate the feasibility of our approach by applying it to the example of developing a driving function concept for a highly automated shuttle.

Keywords: Data Mining and Data Analysis
Abstract: The accuracy of traffic flow prediction significantly impacts the operation of Intelligent Transportation Systems (ITS). In this paper, we propose a Differential Time-variant Traffic Flow Prediction method, which can remarkably improve the accuracy and stability of traffic flow forecast based on deep learning models. First, it is illustrated that importing time information in the form of One-Hot Encoding is effective for traffic flow prediction. Second, data difference is utilized to extract the temporary trend of different locations. This method can better eliminate the uncertainties of traffic flow series like volatility and anomaly. Necessary analyse of our methods are presented to demonstrate the rationality. We combine the two methods together and propose the general structure to exploit them. Three popular deep neural networks are applied to test our method, and experimental results on PeMS data sets indicate that it can make more accurate and stable prediction compared with the same model.

Keywords: Data Mining and Data Analysis, Sensing, Vision, and Perception, Off-line and Online Data Processing Techniques
Abstract: Machine learning components are becoming popular for the automotive industry. More and more data sets become available for training machine learning components. All of them provide ground truth labels for images. The labeling process is expensive and potentially error-prone. At the same time, label correctness defines the business value of a data set. In this paper, we use N-Version approach to assess the label quality in a data set. The approach combines N state-of-the-art neural networks and aggregates their results in a single verdict using majority voting. We analyze this majority vote against the ground truth label and compute the percentage of disagreeing pixels along with other metrics, enabling the automated and detailed analysis of label quality on data sets. We evaluate our methodology by classifying the BDD100K drivable area data set. The evaluation shows that the approach identifies misclassified scenes or inconsistencies between label semantics for similar scenes.

Keywords: Data Mining and Data Analysis, Theory and Models for Optimization and Control, Travel Information, Travel Guidance, and Travel Demand Management
Abstract: Since the introduction of free-floating carsharing (FFCS), system optimization has always been a crucial point in operations. Especially knowledge about the usage of such systems allows for a better understanding, leading to maximized utilization and therefore revenue. In order to understand demand for FFCS services, most often rental data is utilized. However, utilizing such data yields systematic underreporting of demand, since lack of vehicles obstructs counting real demand. In this paper we present an unconstraining algorithm for FFCS system analysis, called Pois_d, that minimizes demand underreporting in rental data due to unavailability. Evaluation of this algorithm shows that it approximates actual demand, reduces underreporting by up to 70% compared to utilizing solely rental data, and reduces error measures by up to 26% as well. Applying Pois_d to real world data, the size of undetected potential in FFCS systems is illustrated. Therefore, the analysis of four areas from the business area of an FFCS provider is presented. Results reveal potential markups on pure rental data of up to 90%. Adjusting demand data for these systems with this algorithm can help to optimize operative measures like vehicle reallocation, adjustment of pricing systems, and planning of business areas.

Keywords: Data Mining and Data Analysis, Off-line and Online Data Processing Techniques, Road Traffic Control
Abstract: Tunnel ventilation systems are strictly controlled by safety regulations. Such regulations define not only their operating conditions during fire situations, but also the way in which they should be activated when the accumulation of pollutant gases reaches certain thresholds that are considered unsafe. In addition to these exceptional circumstances, evacuation of tunnel gases is produced naturally on a regular basis, due to causes like air currents originated in pressure differences among the tunnel portals, or the well known piston effect, as a result of vehicles pushing the air when they pass. This work elaborates on the prediction of air-flow inside the tunnels boosted by traffic flow prediction, in order to assist the system activation, be it automated or manual. After experiments made over real tunnel data with a benchmark of machine learning predictive algorithms,results suggest that traffic flow inside the studied tunnels can be effectively predicted and used to enhance air flow predictions,specially in those cases where an air flow predictor alone is not enough to obtain an actionable forecast. The relevance of these results comes from their direct applicability wherein improving the ventilation activation cycles, by adjusting their automation or by informing operators of future air flow levels.

Keywords: Data Mining and Data Analysis, Driver Assistance Systems, Simulation and Modeling
Abstract: The ability to model and predict ego-vehicle's surrounding traffic is crucial for autonomous pilots and intelligent driver-assistance systems. Acceleration prediction is important as one of the major components of traffic prediction. This paper proposes novel approaches to the acceleration prediction problem. By representing spatial relationships between vehicles with a graph model, we build a generalized acceleration prediction framework. This paper studies the effectiveness of proposed Graph Convolution Networks, which operate on graphs predicting the acceleration distribution for vehicles driving on highways. We further investigate prediction improvement through integrating of Recurrent Neural Networks to disentangle the temporal complexity inherent in the traffic data. Results from simulation with comprehensive performance metrics support that our proposed networks outperform state-of-the-art methods in generating realistic trajectories over a prediction horizon.

Keywords: Driver Assistance Systems, Automated Vehicle Operation, Motion Planning, Navigation, Human Factors in Intelligent Transportation Systems
Abstract: We propose a game theoretic approach to address the problem of searching for available parking spots in a parking lot and picking the "optimal" one to park. The approach exploits limited information provided by the parking lot, i.e., its layout and the current number of cars in it. Considering the fact that such information is or can be easily made available for many structured parking lots, the proposed approach can be applicable without requiring major updates to existing parking facilities. For large parking lots, a sampling-based strategy is integrated with the proposed approach to overcome the associated computational challenge. The proposed approach is compared against a state-of-the-art heuristic-based parking spot search strategy in the literature through simulation studies and demonstrates its advantage in terms of achieving lower cost function values.

Keywords: Driver Assistance Systems, Sensing, Vision, and Perception, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: Autonomous Driving (AD) systems use digital maps as a virtual sensor to perceive the environment around the car. As the field of digital maps continues to evolve, existing solutions face new challenges such as integration ability for new map formats (e.g., High Definition maps), supporting onboard and offboard deployment and providing a generic interface to access the various map data formats. In this paper, we propose a knowledge architecture layer for environmental modeling and distinguish between low-level ontologies based on various map data formats and a high-level ontology for representing a generic road environment. The adequacy of the modeling is validated over two use cases: lane change notification and logical inconsistency detection. The performance is measured using real map data and it shows encouraging results for future development within onboard and offboard systems.

Keywords: Driver Assistance Systems, Advanced Vehicle Safety Systems, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: During the last years' research on Collision Avoidance Systems (CAS) is gaining special attention, due to the decrease of on-road accidents. Current commercial systems can reduce the vehicle speed in case of emergencies such as the appearance of obstacles on the road. However, the behavior of commercial systems is frequently too rigid failing to achieve a proper balance between safety and comfort. In this scenario, this work presents a new approach in which the contextual information of the surrounding environment, such as dedicated infrastructure for vulnerable road users or objects in the vicinity, is used to assess the risks through a Fuzzy inference system. Once risks are evaluated the constraints on the controller acting over the longitudinal vehicle motion are established accordingly. The controller uses a Model Predictive Control (MPC) algorithm. The presented approach illustrates the benefits of modulating the constraints of the MPC controller according to the risk assessment. This approach generates a dynamic speed profile smoothing out critical braking scenarios depending on distances to further objects. For validation, a complex urban scenario was simulated. Results show good performance on the speed planner, also allowing an extendable generalization to different road structures and predefined behaviors from maps and perception systems.

Keywords: Driver Assistance Systems, Sensing, Vision, and Perception, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: Deep neural networks trained end-to-end on demonstrations of human driving have learned to follow roads, avoid obstacles, and take specific turns in intersections to reach a destination. Such conditional imitation learning approach is demonstrated to drive efficiently when deployed on the same training environments, but performance dramatically decreases when deployed to new environments and is not consistent against varying weathers. In this work, a proposed model aims to cope with such two challenges by fusing laser scanner input with the camera. Additionally, a new efficient method of Occupancy Grid Mapping is introduced and used to rectify the model output to further improve the performance. On CARLA simulator urban driving benchmark, the proposed system improves autonomous driving success rate and average distance traveled towards destination on all driving tasks and environments combinations, while it's trained on automatically recorded traces. Autonomous driving success rate generalization improves by 57% and weather consistency improved by around four times.

Keywords: Driver Assistance Systems, Automated Vehicle Operation, Motion Planning, Navigation, Simulation and Modeling
Abstract: Today’s Advanced Driver Assistance Systems (ADAS) adopt an autonomous approach with all instrumentation and intelligence on board of one vehicle. In order to further enhance their benefit, ADAS need to cooperate in the future. This enables, for instance, to solve hazardous situations by coordinated maneuvers for safety intervention on multiple vehicles at the same point in time. Our prototyping environment presented in previous work addresses developing such cooperative ADAS. Its underlying approach is to either bring ideas for cooperative ADAS through the prototyping stage towards plausible candidates for further development, or to discard them as quickly as possible. This is enabled by an iterative process of refining and assessment. In this paper, we focus on handling the application specific parameter space, and more precisely on the scenario related aspects. As a part of our iterative prototyping process, defining and tuning scenarios and application parameters are highly repetitive tasks which needs to be designed very efficiently. We, therefore, strive to create a scenario definition methodology, which provides best flexibility and a minimal expenditure of time on the developer side.

Keywords: Human Factors in Intelligent Transportation Systems, Advanced Vehicle Safety Systems, Sensing, Vision, and Perception
Abstract: The major challenges for analyzing the situation of traffic scenes include defining proper metrics and achieving computation efficiency. This paper proposes two new situation metrics, a multimodality scene model, and a metrics computing network for traffic scene analysis. The main novelty is threefold. (1) The planning complexity and perception complexity are proposed as the situation metrics of traffic senes. (2) A multimodality model is proposed to describe traffic scenes, which combines the information of the static environment, dynamic objects, and ego-vehicle. (3) A deep neural network (DNN) based computing network is proposed to compute the two situation metrics based on scene models. Using the Nuscenes dataset, a high-level dataset for traffic scene analysis is developed to validate the scene model and the situation metrics computing network. The experiment results show that the proposed scene model is effective for situation analysis and the proposed situation metrics computing network outperforms than traditional CNN methods.

Keywords: Human Factors in Intelligent Transportation Systems, Travel Behavior Under ITS
Abstract: Driver drowsiness detection has always been center to research whether for accident risk minimization or recently for driver monitoring in the stages towards automated driving. In this work we analyse videos of visibly alert and less alert drivers collected within a naturalistic driving study in terms of different visual drowsiness metrics. The facial landmark method allows to compute the eye aperture remotely without additional wearables. From this an unsupervised blink detection algorithm is introduced that competes with other supervised methods on benchmark datasets. Common fatigue metrics such as blink rate are considered. We show that there is a significant difference in blink rate between different driver groups and also discuss fatigue levels during the course of a cruise. More importantly, we show that the distribution of eye aperture already displays valuable information on the driver's blinking patterns without the actual need to derive a blink detection system in the first place.

Keywords: Human Factors in Intelligent Transportation Systems, ITS Field Tests and Implementation, Driver Assistance Systems
Abstract: With the growing technological advances in autonomous driving, the transport industry and research community seek to determine the impact that autonomous vehicles (AV) will have on consumers, as well as identify the different factors that will influence their use. Most of the research performed so far relies on laboratory-controlled conditions using driving simulators, as they offer a safe environment for testing advanced driving assistance systems (ADAS). In this study we analyze the behavior of drivers that are placed in control of an automated vehicle in a real life driving environment. The vehicle is equipped with advanced autonomy, making driver control of the vehicle unnecessary in many scenarios, although a driver take over is possible and sometimes required. In doing so, we aim to determine the impact of such a system on the driver and their driving performance. To this end road users’ behavior from naturalistic driving data is analyzed focusing on awareness and diagnosis of the road situation. Results showed that the road features determined the level of visual attention and trust in the automation. They also showed that the activities performed during the automation affected the reaction time to take over the control of the vehicle.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Automated Vehicle Operation, Motion Planning, Navigation, Driver Assistance Systems
Abstract: Reinforcement learning is one of the methods that has been used to realize optimal driving. Most studies have focused on evaluating learning performance of a fraction of vehicles controlled by reinforcement learning. It is unclear how these controlled vehicles influence other vehicles. We conducted several experiments examining the impact of multiple vehicles controlled by reinforcement learning on traffic flow. The simulations were performed on a three-lane freeway with lane regulation at the end of one of the lanes. The controlled vehicles were trained to drive as fast as possible and run non-cooperatively. We found out that controlled vehicles could run faster than human-driven vehicles. Moreover, we anticipated that if multiple vehicles were run selfishly, it would adversely affect traffic flow. Contrary to expectations, the experimental results showed that even if numerous controlled vehicles drive selfishly, the negative impact on overall traffic would be small.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Automated Vehicle Operation, Motion Planning, Navigation, Air Traffic Management
Abstract: We present a solution to the problem of fairly planning a fleet of Unmanned Aerial Vehicles (UAVs) that have different missions and operators, such that no one operator unfairly gets to finish its missions early at the expense of others - unless this was explicitly negotiated. When hundreds of UAVs share an urban airspace, the relevant authorities should allocate corridors to them such that they complete their missions, but no one vehicle is accidentally given an exceptionally fast path at the expense of another, which is thus forced to wait and waste energy. Our solution, FairFly, addresses the fair planning question for general autonomous systems, including UAV fleets, subject to complex missions typical of urban applications. FairFly formalizes each mission in temporal logic. An offline search finds the fairest paths that satisfy the missions and can be flown by the UAVs, leading to lighter online control load. It allows explicit negotiation between UAVs to enable imbalanced path durations if desired. We present three fairness notions, including one that reduces energy consumption. We validate our results in simulation, and demonstrate a lighter computational load and less UAV energy consumption as a result of flying fair trajectories.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems
Abstract: Motion prediction of surrounding vehicles is one of the most important tasks handled by a self-driving vehicle, and represents a critical step in the autonomous system necessary to ensure safety for all the involved traffic actors. Recently a number of researchers from both academic and industrial communities have focused on this important problem, proposing ideas ranging from engineered, rule-based methods to learned approaches, shown to perform well at different prediction horizons. In particular, while for longer-term trajectories the engineered methods outperform the competing approaches, the learned methods have proven to be the best choice at short-term horizons. In this work we describe how to overcome the discrepancy between these two research directions, and propose a method that combines the disparate approaches under a single unifying framework. The resulting algorithm fuses learned, uncertainty-aware trajectories with lane-based paths in a principled manner, resulting in improved prediction accuracy at both shorter- and longer-term horizons. Experiments on real-world, large-scale data strongly suggest benefits of the proposed unified method, which outperformed the existing state-of-the-art. Moreover, following offline evaluation the proposed method was successfully tested onboard a self-driving vehicle.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Driver Assistance Systems, Cooperative Techniques and Systems
Abstract: Platooning is now a hot research topic due to its advantages in traffic efficiency and fuel economy. This paper investigates the internal stability of vehicular platoons with communication topology disturbances. An inertially delayed linear vehicle model is employed. Communication topologies are modeled by directed graphs. Using a distributed linear feedback controller, we yield a high dimensional linear model for the closed-loop platoon dynamics. Then a Riccati inequality based algorithm is proposed to calculate the stabilizing control gain. It is proven that a lower bound on the dwell time of the topology change should be satisfied, so that asymptotic platoon stability can be achieved. Finally, the proposed theorem is validated through numerical simulations.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Electric Vehicles, Emission and Noise Mitigation
Abstract: Urban traffic congestion is a chronic problem faced by many cities in the US and worldwide. It results in inefficient infrastructure use as well as increased vehicle fuel consumption and emission levels. Excessive fuel consumptions add extra costs to commuters as well as transportation businesses. Consuming less fuel and thus reducing costs by a single percentage digit can have a significant impact on the balance sheet as well as the protection of the environment. Researchers have developed, and continue to develop, tools and systems to optimize the operations of fleets as well as engines in order to burn less fuel and therefore generate less CO2 emissions. Platooning is one such tool that attempts to maintain relatively small distances (i.e. pre-determined time gap) between consecutive vehicles. It has the potential to increase the capacity of the road as well as reduce the consumed fuel. In this paper, we use a fuel consumption model for internal combustion light-duty vehicles, electric vehicles, hybrid electric vehicles, buses and trucks in order to determine and quantify the effects of platooning on a fleet fuel consumption. The results suggest that a reduction of up to 3%, 3.5%, 4.5 %, 10%, and 15% in fuel consumption can be achieved for internal combustion engine vehicles, hybrid electric vehicles, electric vehicles, buses and trucks, respectively.

Keywords: Sensing, Vision, and Perception
Abstract: In this paper, we propose a novel form of the loss function to increase the performance of LiDAR-based 3D object detection and obtain more explainable and convincing uncertainty for the prediction. The loss function was designed using corner transformation and uncertainty modeling. With the new loss function, the performance of our method on the val split of KITTI dataset shows up to a 15% increase in terms of Average Precision (AP) comparing with the baseline using simple L1 Loss. In the study of the characteristics of predicted uncertainties, we find that generally more accurate prediction of the bounding box is accompanied by lower uncertainty. The distribution of corner uncertainties agrees on the distribution of the point cloud in the bounding box, which means the corner with denser observed points has lower uncertainty. Moreover, our method learns the constraint from the cuboid geometry of the bounding box in the uncertainty prediction. Finally, we propose an efficient Bayesian updating method to recover the uncertainty for the original parameters of the bounding boxes which can help provide probabilistic results for the tracking and planning module.

Keywords: Sensing, Vision, and Perception, Cooperative Techniques and Systems, Off-line and Online Data Processing Techniques
Abstract: Ground truth data plays an important role in validating perception algorithms and in developing data-driven models. Yet, generating ground truth data is a challenging process, often requiring tedious manual work. Thus, we present a post-processing approach to automatically generate ground truth data from environment sensors. In contrast to existing approaches, we incorporate raw sensor data from multiple vehicles. As a result, our cooperative fusion approach overcomes drawbacks of occlusions and decreasing sensor resolution with distance. To improve the alignment precision for raw sensor data fusion, we include mutual detections and match the jointly-observed static environment to support differential global positioning system localization. We further provide a new registration algorithm, where all point clouds are moved simultaneously, while restricting the transformation parameters to increase the robustness against misalignments. The benefits of our raw sensor data fusion approach are demonstrated with real lidar data from two test vehicles in different scenarios.

Keywords: Sensing, Vision, and Perception, Communications and Protocols in ITS, Other Theories, Applications, and Technologies
Abstract: Recently, learned image compression by means of neural networks has experienced a performance boost by the use of adversarial loss functions. Typically, a generative adversarial network (GAN) is designed with the generator being an autoencoder with quantizer in the bottleneck for compression and reconstruction. It is well known from rate-distortion theory that vector quantizers provide lower quantization errors than scalar quantizers at the same bitrate. Still, learned image compression approaches often use scalar quantization instead. In this work we provide insights into the image reconstruction quality of the often-employed uniform scalar quantizers, non-uniform scalar quantizers, and the rarely employed but bitrate-efficient vector quantizers, all being integrated into backpropagation and operating under the exact same bitrate. Further interesting insights are obtained by our investigation of an MSE loss and a GAN loss. We show that vector quantization is always beneficial for the compression performance both in the latent space and the reconstructed image space. However, image samples demonstrate that the GAN loss produces the more pleasing reconstructed images, while the non-adversarial MSE loss provides better quality scores of various instrumental measures both in the latent space and on the reconstructed images.

Keywords: Sensing, Vision, and Perception, Driver Assistance Systems
Abstract: Environment modeling utilizing sensor data fusion and object tracking is crucial for safe automated driving. In recent years, the classical occupancy grid map approach, which assumes a static environment, has been extended to dynamic occupancy grid maps, which maintain the possibility of a low-level data fusion while also estimating the position and velocity distribution of the dynamic local environment. This paper presents the further development of a previous approach. To the best of the author’s knowledge, there is no publication about dynamic occupancy grid mapping with subsequent analysis based only on radar data. Therefore in this work, the data of multiple radar sensors are fused, and a grid-based object tracking and mapping method is applied. Subsequently, the clustering of dynamic areas provides high-level object information. For comparison, also a lidar-based method is developed. The approach is evaluated qualitatively and quantitatively with real-world data from a moving vehicle in urban environments. The evaluation illustrates the advantages of the radar-based dynamic occupancy grid map, considering different comparison metrics.

Keywords: Sensing, Vision, and Perception, Aerial, Marine and Surface Intelligent Vehicles, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: Environmental perception is one of the critical aspects of autonomous driving for maritime applications, especially in fields of self-navigation and maneuver planning. For near-field recognition, this paper proposes a novel framework for multi-sensor data fusion, which is able to determine the occupied static space and to track dynamic objects simultaneously. An unmanned surface vessel (USV) is equipped with LiDAR sensors, a GNSS receiver, and an Inertial Navigation System (INS). In the framework, the point cloud from LiDAR sensors is firstly clustered into various objects, then associated with known objects in the past. After dynamic segmentation, the static objects are represented using an optimized occupancy grid map and the dynamic objects are tracked and matched to corresponding Automatic Identification System(AIS) messages. The proposed algorithms are validated with data collected from real-world tests, which are conducted in the Rostock Harbor, Germany. After applying the proposed algorithm, the perceived test area can be represented with a 3D occupancy grid map with a 10,cm resolution. At the same time, dynamic objects in the view are detected and tracked successfully with an error of less than 10 %. The plausibility of the results is qualitatively evaluated by comparing with Google Maps and the corresponding AIS messages.

Keywords: Simulation and Modeling, Other Theories, Applications, and Technologies
Abstract: Recently, reinforcement learning (RL) has been used as a tool for finding failures in autonomous systems. During execution, the RL agents often rely on some domain-specific heuristic reward to guide them towards finding failures, but constructing such a heuristic may be difficult or infeasible. Without a heuristic, the agent may only receive rewards at the time of failure, or even rewards that guide it away from failures. For example, some approaches give rewards for taking more likely actions, in order to to find more likely failures. However, the agent may then learn to only take likely actions, and may not be able to find a failure at all. Consequently, the problem becomes a hard-exploration problem, where rewards do not aid exploration. A new algorithm, go-explore (GE), has recently set new records on benchmarks from the hard-exploration field. We apply GE to adaptive stress testing(AST), one example of an RL-based falsification approach that provides a way to search for the most-likely failure scenario. We simulate a scenario where an autonomous vehicle drives while a pedestrian is crossing the road. We demonstrate that GE is able to find failures without domain-specific heuristics, such as the distance between the car and the pedestrian, on scenarios that other RL techniques are unable to solve. Furthermore, inspired by the robustification phase of GE, we demonstrate that the backwards algorithm (BA) improves the failures found by other RL techniques.

Keywords: Simulation and Modeling, Network Modeling
Abstract: Simulation is a valuable building block for the verification and validation of automated driving functions (ADF). When simulating urban driving scenarios, simulation maps are one important component. Often, the generation of those road networks is a time consuming and manual effort. Furthermore, typically many variations of a distinct junction or road section are demanded to ensure that an ADF can be validated in the process of releasing those functions to the public.

Therefore, in this paper, we present a prototypical solution for a logical road network description which is easy to maintain and modify. The concept aims to be non-redundant so that changes of distinct quantities do not affect other places in the code and thus the variation of maps is straightforward. In addition, the simple definition of junctions is a focus of the work. Intersecting roads are defined separately, are then set in relation and the junction is finally generated automatically.

The idea is to derive the description from a commonly used, standardized format for simulation maps in order to generate this format from the introduced logical description. Consequently, we developed a command-line tool that generates the standardized simulation map format OpenDRIVE.

Keywords: Simulation and Modeling, Theory and Models for Optimization and Control, Traffic Theory for ITS
Abstract: This work contributes to the operational design of dynamic HOV lanes on French urban freeways. We propose a framework to estimate the acceptable dynamic maximal speed of the HOV lane, knowing the current origin speed. The aim is to ensure safety of lane changes while maintaining an acceptable duration of the period before gap acceptance. It is recognized that the longer the time a driver waits before accepting a gap, the lower he/she places the threshold for next gap acceptance. That may lead to inappropriate behaviors, such as unsafe lane changes.

The key variables for lane changing realization are speeds and headways on origin and target lanes (estimated here with individual loop data). Both variables play an important role in lane change maneuver safety. We present in this paper two lane change realization models combining simple rules and observed data, one pessimistic and one optimistic. From this, we determine the maximal and minimal boundaries of two indicators: the number of rejected gaps and the waiting time to safe lane change. Those indicators lead to an appraisal of lane change realization safety.

The results tend to show that a speed regulation lower than usual speed may increase safety while not degrading the service provided by this lane to its specific users. Having in mind the perspective of lanes dedicated to connected automated cruise control vehicles, we extend the proposed method to provide an analysis of the impact of uniforms inter-vehicular time gap on lane change safety in this case.

Keywords: Simulation and Modeling, Automated Vehicle Operation, Motion Planning, Navigation, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: The multiple vehicle routing problem (mVRP) concerns the scheduling of multiple vehicles so as to visit some locations of interest. We study a dynamic version of mVRP where the travel costs are not a priori known and may vary at runtime. Moreover, we introduce energy-related constraints which make the problem more complex. Vehicles have only finite energy reserves, which gradually diminish as they move between different locations, but can also gain some energy at specific depot locations. The objective is to visit all locations of interest as fast as possible without any vehicle exhausting its energy. We propose an online algorithm based on the Large Neighbourhood Search (LNS) heuristic. We evaluate the algorithm for different topologies and degrees of vehicle autonomy. Our results show that it achieves significantly better results than an offline algorithm that produces a safe schedule based on worst-case cost estimates.

Keywords: Simulation and Modeling, Theory and Models for Optimization and Control, Travel Information, Travel Guidance, and Travel Demand Management
Abstract: The estimation of travel time distribution (TTD) is critical for reliable route guidance and provides theoretical bases and technical support for advanced traffic management and control. The state-of-the art procedure for estimating arterial TTD commonly assumes that the path travel time follows a certain distribution without considering segment correlation. However, this approach is usually unrealistic as travel times on successive segments may be dependent. In this study, copula functions are used to model arterial TTD as copulas are able to incorporate for segment correlation. First, segment correlation is empirically investigated using day-to-day GPS data provided by BMW Group for one major urban arterial in Munich, Germany. Segment TTDs are estimated using a finite Gaussian Mixture Model (GMM). Next, several copula models are introduced, namely Gaussian, Student-t, Clayton, and Gumbel, to model the dependent structure between segment TTDs. The parameters of each copula model are obtained by Maximum Log Likelihood Estimation. Then, path TTDs comprised of consecutive segment TTDs are estimated based on the copula models. The scalability of the model is evaluated by investigating the performance for an increasing number of aggregated links. The best fitting copula is determined in terms of goodness-of-fit test. The results demonstrate the advantage of the proposed copula model for an increasing number of aggregated segments, compared to the convolution without incorporating segment correlations.

Keywords: Theory and Models for Optimization and Control, Automated Vehicle Operation, Motion Planning, Navigation, Other Theories, Applications, and Technologies
Abstract: In this paper, we propose an algorithm for optimal generation of nonholonomic paths for planning parking maneuvers with a kinematic car model. We demonstrate the use of Successive Convexification algorithms (SCvx), which guarantee path feasibility and constraint satisfaction, for parking scenarios. In addition, we formulate obstacle avoidance with state-triggered constraints which enables the use of logical constraints in a continuous formulation of optimization problems. This paper contributes to the optimal nonholonomic path planning literature by demonstrating the use of SCvx and state-triggered constraints which allows the formulation of the parking problem as a single optimization problem. The resulting algorithm can be used to plan constrained paths with cusp points in narrow parking environments.

Keywords: Theory and Models for Optimization and Control, Road Traffic Control
Abstract: In our previous work, we proposed a Virtual Phase-Link street traffic model to provide optimal control of green splits. The simulations implemented the offsets which were obtained from Vistro offline. Offsets can significantly impact the performance of arterial traffic controls. This paper introduces offsets as optimization variables to the Virtual Phase-Link street traffic model. Based on the optimization results from the optimal green splits control proposed in the previous chapters, we derived the delay function for offsets optimization. The proposed offsets optimization was tested under two scenarios of the same arterial against their base cases in simulations. It was found that in both scenarios, the proposed method resulted in significantly less delay compared to the base cases. It was also found that the proposed offsets optimization method can identify the dominant traffic path and provide progression optimization for it.

Keywords: Theory and Models for Optimization and Control, Data Mining and Data Analysis, Traffic Theory for ITS
Abstract: New smart traffic signal control algorithms are capable of predicting the traffic signal state (red, green, or amber) changes, which can be provided to users to achieve more efficient traffic flow. However, these predictions pose an uncertain impact on the traffic flow and safety depending upon the quality of prediction. The information regarding the current state as well as the predicted residual time of state is communicated to other users in the form of Signal Phase and Timing (SPaT) data. In this paper, the SPaT message data is analyzed from an on-road pilot of different traffic signal control algorithms on provincial roads in the Province of North Holland. For analysis, new methods and indicators for quantification of prediction accuracy and quality of algorithm are proposed. These indicators can either be used for correction of state change prediction in real-time or for comparative analysis of the performance of different traffic signal control algorithms. This paper presents three main findings. First, it is found that a half fixed algorithm has very high prediction accuracy up to 99% for optimized directions. Second, the prediction accuracy of Time to Amber predictions improved by around 30% with this algorithm. Third, the overall reliability of prediction always increased with the use of the algorithm.

Keywords: Theory and Models for Optimization and Control, Cooperative Techniques and Systems, Sensing, Vision, and Perception
Abstract: Traffic signals provide one of the primary means to administer conflicting traffic flows. Existing signal control strategies, operating on hand-crafted rules, fail to efficiently, autonomously adapt to the changing traffic patterns. Each signal control system independently manages one intersection at a time and regulates navigation of vehicles through that intersection. Current systems cannot co-operate to optimize aggregate traffic flows through multiple road intersections. Consequently, they are susceptible to making myopic signal control decisions that might be effective locally, but not globally. Instead, we propose a system of multiple, coordinating traffic signal control systems. This paper presents the first application of multi-agent deep reinforcement learning (DRL) to achieve traffic optimization through multiple road intersections solely based on raw pixel input from CCTV cameras in real time. This set of traffic control agents is shown to significantly outperform independently operating (both DRL-trained and loop-induced) adaptive signal control systems, by increasing traffic throughput and reducing the average time a vehicle spends in an intersection. Additionally, this paper, introduces attention-based visualization to interpret and validate the proposed multi-agent signal control methodology.

Keywords: Theory and Models for Optimization and Control, Intelligent Logistics, Commercial Vehicle Electronic Clearance
Abstract: Contemporary electronic marketing leads to massive requirements of courier services in China. A local outlet providing delivery and real-time pickup services, however, severely depends on the good staff of experience to handle the routing tasks. These routing tasks are formulated as a one-to-many-to-one dynamic pickup and delivery problem.

In this research, we have developed an online method to solve routing problems. With adaptive memory and heuristic insertion for a speedy response, this method generates results by taking both quality and responsiveness into account, based on simulated annealing to optimize untravelled routes during the trip. This real-time method enables to establish a real-time route planning system: after initialization, adaptive memory is built up to contain the multiple candidate solutions and updated by real-time optimization responding to real-time requests insertion; once a dispatching order is needed, the best solution from the adaptive memory will be selected.

By testing on simulated data with different dynamism level, we have gained good results of both better responsiveness and quality than that of the greedy algorithm, and showing that data with high dynamism can also have low-cost solutions. This work contributes to reducing human involvement in real-time courier service.

Keywords: Communications and Protocols in ITS, Cooperative Techniques and Systems, ITS Field Tests and Implementation
Abstract: EU-China 5G collaboration trials will be conducted addressing two specific scenarios: (1) enhanced Mobile Broadband (eMBB) on the 3.5GHz band; and (2) Internet of Vehicles (IoV) based on LTE-V2X using the 5.9 GHz band for Vehicle-to-Vehicle (V2V) and the 3.5 GHz band for Vehicle-to-Network (V2N). This paper discussed scenario 2, and presents the use cases based on next generation communication technologies in the domain of Cooperative Intelligent Transport Systems (C-ITS) for cooperative and automated road transport. In addition, it describes for each use case the developed physical architecture. Finally, it provides an overview of the joint V2X trials to be conducted in the EU and China, in the context of the 5G-DRIVE project.

Keywords: Communications and Protocols in ITS, Network Management, Simulation and Modeling
Abstract: The vehicle density determines the frequency of information congestion and collision in VANETs, and affects the quality of communication. Cluster management of vehicle nodes can effectively improve communication efficiency of the network. A clustering framework based on changes in vehicle density and an improved k-means clustering algorithm based on vehicle movement characteristics are proposed in this paper. According to the change of vehicle density, nodes are dynamically clustered, in some cases MSCNs(Mobile Secondary Computing Node) are selected and virtual computing areas are divided. The simulation on NS3 shows that the proposed improved kmeans- based adaptive clustering algorithm has strong stability and high communication efficiency.

Keywords: Communications and Protocols in ITS, Sensing and Intervening, Detectors and Actuators, Data Mining and Data Analysis
Abstract: Technology development in the field of the Internet of Things (IoT) and more specifically in Low-Power Wide-Area Networks (LPWANs) has enabled a whole set of new applications in several fields of Intelligent Transportation Systems. Among all, smart-railways represents one of the most challenging scenarios, due to its wide geographical distribution and strict energy-awareness. This paper aims to provide an overview of the state-of-the-art in LPWAN, with a focus on intelligent transportation. This study is part of the RAILS (Roadmaps for Artificial Intelligence integration in the raiL Sector) research project, funded by the European Union under the Shift2Rail Joint Undertaking. As a first step to meet its objectives, RAILS surveys the current state of development of technology enablers for smart-railways considering possible technology transfer from other sectors. To that aim, IoT and LPWAN technologies appear as very promising for cost-effective remote surveillance, monitoring and control over large geographical areas, by collecting data for several sensing applications (e.g., predictive condition-based maintenance, security early warning and situation awareness, etc.) even in situations where power supply is limited (e.g., where solar panels are employed) or absent (e.g., installation on-board freight cars).

Keywords: Communications and Protocols in ITS, Other Theories, Applications, and Technologies, Simulation and Modeling
Abstract: LTE for Railway (LTE-R) is a promising railway mobile communication system to provide new communicationbased applications for smart railways, but it is plagued by inter-cell co-channel interference. In order to quantify the impact of inter-cell interference, it is necessary to obtain the downlink Signal to Interference plus Noise Ratio (SINR) distribution for system performance evaluation. The singlecarrier SINR is analyzed based on LogNormal (LN) and Log SKew Normal (LSKN) approximations. Further, using Exponentially Effective SINR Mapping (EESM) method, the single-carrier SINR analysis results are extended to the multicarrier LTE-R system. A closed-form expression for the effective SINR is derived by approximating it as a LN distribution. Based on the above analysis results, the impacts of overlapping depth between neighboring cells on the outage probability and capacity are investigated. Our theoretical analysis is verified with high accuracy by Monte Carlo simulation. The simulation results show that the overlapping depth has a weak effect on the performance of center users, but deeply overlapped coverage leads to a serious deterioration in outage performance of edge users.

Keywords: Communications and Protocols in ITS, Advanced Vehicle Safety Systems, Simulation and Modeling
Abstract: Fault detection and diagnosis of high-speed train (HST) plays a pivotal role in intelligent rail transportation system. It aims at making real-time operation decisions based on evaluation results of train states to guarantee the operation safety. Thanks to the rapid development of edge computing technology, the performance of fault detection and diagnosis application can be improved dramatically through offloading computing tasks. In this paper, the fault detection and diagnosis service of HST is modeled as a vehicular edge computing (VEC) application where the application is partitioned into multiple tasks. Then, the edge computing model of fault detection and diagnosis is proposed, in which wireless transmission models and cost models are modeled. Based on the proposed application's models, the task offloading problem on minimizing the execution cost is formulated. Then, an offloading strategy with minimum objective cost is proposed. The effectiveness of the proposed offloading strategy is validated by extensive simulation results. The simulation results show that the execution costs can be reduced up to about 70%.

Keywords: Incident Management, Management of Exceptional Events: Incidents, Evacuation, Emergency Management, Data Mining and Data Analysis
Abstract: Forecasting short-term, nonrecurring traffic dynamics caused by incidents is an essential capability in the Intelligent Transportation Systems. This research proposes a prediction framework in which Conditional Deep Convolutional Generative Adversarial Nets (C-DCGAN) is trained to predict the traffic spillbacks patterns associated with freeway incidents at merging bottleneck. Speed tensors, which depict the spatiotemporal incident-induced impacts for multiple neighboring routes, is a suitable object for the GAN model to understand and predict. Further, we demonstrated how to use the mesoscopic Dynamic Traffic Assignment (DTA) model DynusT to generate a large number of training data, thus speeding up the model training. The developed model achieves both statistical and spatial similarities between predicted speed tensors and actual tensors, to 83.84%. To the best of our knowledge, this line of work is one of the first attempts in the literature to train the Machine Learning model to predict speed tensor representation of multi-location incident-induced spatiotemporal impact at merging bottleneck and speeding up the training via simulation.

Keywords: Incident Management, Data Mining and Data Analysis, Management of Exceptional Events: Incidents, Evacuation, Emergency Management
Abstract: Compared to normal incidents, secondary incidents are more likely to result in severe injuries and fatalities. However, limited efforts have been made to unveil the factors affecting the severity of secondary incidents. Incidents that occurred on the Interstate-5 in California within five years were collected. Detailed real-time traffic flow conditions, geometric characteristics, and weather conditions were obtained. First, a Random Forest-based (RF) feature selection approach was adopted. Then, Support Vector Machine (SVM) models were developed to investigate the effects of contributing factors. For comparison, RF and Ordered Logistic (OL) models were also built based on the same dataset. It was found that the SVM model has a high capacity for solving classification problems with limited data availability. Further, sensitivity analysis assessed the impacts of explanatory variables on the injury severity level. Explanatory variables, including occupancy, duration, frequency of lanes changes, and number of lanes, were found to contribute to injury severity of secondary incidents. Smoothing these traffic conditions after an incident occurs and responding fast in incident handling and clearance have the potential to reduce road trauma caused by secondary incidents.

Keywords: Incident Management, Aerial, Marine and Surface Intelligent Vehicles, Infrastructure for Charging, Communication and Controls
Abstract: With the recent development of Unmanned Aerial Vehicles (UAV) applications, traffic police might utilize UAV to conduct Service Patrol (SP) tasks. However, a major limitation of existing UAV systems is their limited flight endurance. To address this issue, by implementing the auto-rechargeable mechanism, we explicitly optimize hardware setting and system strategy required for regional SP with predefined initial tasks and stochastic incidents by solving a heuristic facility location problem and multi-objective path planning problem based on cooperative auto-recharging facilities, and fleet management center. The proposed fleet size and system performance are leveraged in a grid network with respect to different infrastructure settings and service coverage. The field experiments were conducted in Xi’an for SP tasks in complete vehicle coverage trajectory reconstruction, and results show that the proposed system is capable of unmanned SP tasks and large-scale application in urban scenarios.

Keywords: Management of Exceptional Events: Incidents, Evacuation, Emergency Management
Abstract: Disasters of different types are an increasing threat to modern societies. The speed at which people can evacuate a disaster may mean the difference between life or death. One way of modelling disaster evacuation is by using multi-agent simulation, in which each person in a population is simulated as an autonomous individual. This enables emergency management groups to analyse the results across a range of scenarios, under varying parameter settings, to better understand how to reduce evacuation times. Previous studies have used this approach to model disaster evacuations, but only considered evacuation on foot. This research studies the effect of public transport on the evacuation process using the MultiAgent Transport Simulation (MATSim) platform. Experimental results show large reductions in the average evacuation time of agents can be achieved using public transport, compared with pedestrian traffic only. A sensitivity analysis illustrates the conditions under which additional buses can be added to further reduce evacuation time and the limit of this improvement.

Keywords: Management of Exceptional Events: Incidents, Evacuation, Emergency Management, Road Traffic Control, Travel Information, Travel Guidance, and Travel Demand Management
Abstract: Abstract— In view of the fact that existing metrics such as reliability and vulnerability cannot effectively and comprehensively describe road network performances under sudden major disruptive events, a connectivity resilience metric together with two resilience-based component importance measures are proposed. Bayesian network (BN) is served as a modelling tool which is used to assess the road network resilience and component importance under different earthquake magnitudes, and a case study is conducted on the Nguyen and Dupuis network. The results show that the higher the earthquake level, the lower the system resilience. The road network resilience can be deduced from links resilience and critical links can be found by BN. Two resilience-based link importance measures can effectively rank the importance of links, and the importance ranks of most links change over time. All these conclusions verify the feasibility of the proposed model. Key words— connectivity resilience; Bayesian network; urban road networks; probability importance measure; criticality importance measure; earthquake

Keywords: Management of Exceptional Events: Incidents, Evacuation, Emergency Management, Off-line and Online Data Processing Techniques, Data Mining and Data Analysis
Abstract: Traffic flow data collected by loop detectors have been widely used for traffic incident detection. As traffic flow data have strong spatial-temporal correlations, this study tries to detect traffic incidents using an unsupervised learning approach. In this paper, a novel automatic incident detection (AID) method based on Autoencoder (AE) is proposed to detect the occurrence time and the location of traffic incidents in both freeway and urban networks. AE is an unsupervised machine learning model, which extracts nonlinear features of traffic flow data. A statistic named Squared Prediction Error (SPE) is constructed for incident detection. Meanwhile, the contribution plot technique is applied for incident localization. The experiments are conducted via a microscopic simulation platform Vissim and the test results verify the timeliness, effectiveness, and transferability of the proposed method.

Keywords: Emergency Vehicle Management, Travel Behavior Under ITS, Simulation and Modeling
Abstract: Reducing the response time of Emergency Vehicles (EVs) has an undoubted advantage in saving life and property. Implementing pre-emption can aid in achieving it. EVs get unobstructed movement via pre-emption, usually by altering traffic signals and giving a green wave throughout the route. This approach of absolute pre-emption effects adversely on regular traffic by imposing unnecessary waiting. In this paper, we propose a novel emergency vehicle pre-emption (EVP) algorithm implemented in the Vehicular Ad-hoc Network (VANET) that can reduce the imposed undesirable waiting time, but still ascertains EVs meet target response time. We introduce mixed-criticality real-time system scheduling concept where different level of emergencies is mapped with different criticality levels and assign certain success assurance level to respective criticality. We implemented the EVP algorithm for an arterial traffic network and leveraged the use of valuable information that gets transmitted via VANET to make critical decisions. The proposed algorithm can significantly reduce the average waiting time of regular traffic. It also ascertains all EVs with different level of criticality meet target response time respective to their assurance level.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation
Abstract: This paper presents a novel path planning algorithm for Autonomous Land Vehicles (ALVs), which makes two significant improvements to the traditional A-star algorithm. An evaluation standard is first introduced to measure the performance of different algorithms and to select appropriate parameters for the proposed algorithm. Then, a guide-line based A-star algorithm is presented, in which the guide-line is employed to develop the heuristic function to overcome the shortcoming of traditional A-star algorithms. Further, for improving the obstacle avoidance performance, a novel key-point based algorithm is presented, which would guide the planning path to avoid the obstacle much earlier than the traditional one. Combination of these two improvements, this improved A-Star based path planning algorithm is valid and lots of experiments are carried out. Experimental results show that the performance of the proposed algorithm is robust and stable. Compared with the state-of-the-art techniques, the performance is better in both simulation and real application.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Sensing, Vision, and Perception, Advanced Vehicle Safety Systems
Abstract: In an end-to-end vehicle control scenario, where a deep neural network is trained on visual input solely, adversarial vulnerability leaves a possibility to manipulate the steering predictions. Patch-based adversarial attacks present an especially serious menace, because they can be performed in the real world by printing out a generated universal pattern. However, the boundary conditions and feasibility of such attacks to compromise the security of autonomous vehicles have been only sparsely studied so far.

We demonstrate and evaluate such attacks in the CARLA simulative environment under different weather and lighting settings, while conducting experiments in open and closed loop attack scenarios. Our findings reveal that attack strength is highly dependent on the surrounding location as well as on environment conditions. We also observe that attack success in an open loop scenario only partially coincides with that in a closed loop scenario. This analysis helps to set the stage for future experiments on public roads.

Furthermore, we propose a defense concept to remove malignant perturbations from an input image, which does not affect its salient regions. We analyze deviations from the unattacked vehicle trajectory both on adversarial and suppressed inputs.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Theory and Models for Optimization and Control
Abstract: Maneuvering in dense traffic is a challenging task for autonomous vehicles because it requires reasoning about the stochastic behaviors of many other participants. In addition, the agent must achieve the maneuver within a limited time and distance. In this work, we propose a combination of reinforcement learning and game theory to learn merging behaviors. We design a training curriculum for a reinforcement learning agent using the concept of level-k behavior. This approach exposes the agent to a broad variety of behaviors during training, which promotes learning policies that are robust to model discrepancies. We show that our approach learns more efficient policies than traditional training methods.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Driver Assistance Systems
Abstract: A new multi-model recurrent neural network (RNN) control scheme is developed for autonomous vehicle lane-change maneuvering with longitudinal speed variation. Lateral motion control for lane-change maneuvering under longitudinal speed variation becomes challenging because the lateral vehicle dynamics is very involved. The literature has studied lane-change control using a bicycle dynamic model with fixed longitudinal speed. However, It rarely reported how a lane-change controller under variation of speed performs. In the paper, we develop an innovative scheme in which multiple RNNs are trained. And a probabilistic data association of their outputs is given as the command to the steering angle. Each RNN is trained by optimizing the corresponding model predictive control (MPC) with fixed vehicle speed. Further, the discrete probability distribution is used to avoid impractical RNN training for lane-change maneuvering with various vehicle speed variation scenarios. The proposed multi-model RNN control scheme is demonstrated through an application. The proposed system shows that it satisfies the constraints given in the design of MPCs and exhibit better control performance.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Human Factors in Intelligent Transportation Systems, Driver Assistance Systems
Abstract: In this paper, we propose a vehicle longitudinal control with a velocity profile for stop and go (SG) operation. A three-phase velocity profile is proposed to achieve smooth velocity control for stop and go operation. The proposed velocity profile makes the vehicle stop at the desired stationary point without overshooting. Given the maximum deceleration limit, we design a velocity profile for the first and second phases until the vehicle passes a setpoint near before the stop point. Once the vehicle passes the setpoint, the car is controlled by a position controller to make a complete stop at the stop point. The three-phase velocity profile considers the maximum jerk, the maximum acceleration, time to collision, and distance to go. With the proposed method, we can achieve the SG operation, providing a comfortable ride to passengers. The proposed SG operation method is compared with conventional adaptive cruise control. From the computational study, we confirm that the proposed method shows satisfactory performances for various initial speed and distance to the stationary point with CarSim and MATLAB/Simulink.

Keywords: Data Mining and Data Analysis, Advanced Vehicle Safety Systems, Simulation and Modeling
Abstract: In this paper, we explore different Convolutional Neural Network (CNN) architectures to extract features in a Time to Lane Change (TTLC) classification problem for highway driving functions. These networks are trained using the HighD dataset, a public dataset of realistic driving on German highways. The investigated CNNs achieve approximately the same test accuracy which, at first glance, seems to suggest that all of the algorithms extract features of equal quality. We argue however that the test accuracy alone is not sufficient to validate the features which the algorithms extract. As a form of validation, we propose a two pronged approach to confirm the quality of the extracted features. In the first stage, we apply a clustering algorithm on the features and investigate how logical the feature clusters are with respect to both an external clustering validation measure and with respect to expert knowledge. In the second stage, we use a state-of-theart dimensionality reduction technique to visually support the findings of the first stage of validation. In the end, our analysis suggests that the different CNNs, which have approximately equal accuracies, extract features of different quality. This may lead a user to choose one of the CNN architectures over the others.

Keywords: Data Mining and Data Analysis, Driver Assistance Systems, Advanced Vehicle Safety Systems
Abstract: With the increasing prevalence of autonomous driving, research on traffic sign detection (TSD) draws substantial attention recently. Existing studies usually adopt a two-step framework by first enhancing features of small Region of Interests and then for image analytic. However, the training process of this approach can be in-stable due to the lack of context information, which may restrict the quality of super-resolution features. In this paper, we propose an efficient one-step learning-based solution to deal with the TSD problem. We first develop an improved Faster RCNN to detect small objects in traffic images. Then we introduce a new sampling method to optimize the proposed network by selecting high-quality proposals. We also present a post-processing scheme to resample the hard false samples having significant contributions to network optimization. Moreover, we adopt Res2net as the backbone of the proposed network in order to obtain more discriminative features. We conduct extensive experiments on the Tsinghua-Tencent 100k dataset, and the results show that our method outperforms other algorithms in terms of accuracy and recall.

Keywords: Data Mining and Data Analysis, Travel Information, Travel Guidance, and Travel Demand Management, Road Traffic Control
Abstract: This paper introduces a methodology which identifies congestion hot spots for individual congestion types. The proposed algorithm first isolates coherent congested clusters out of a spatio-temporally discretized speed matrix. Then, virtually driven trajectories which pass through the respective congestion area are calculated and their speed profiles are analyzed. A congestion type is assigned to each trajectory and thereafter, a congestion type for the overall cluster is determined. Considering the spatial and temporal start and end points of each cluster along with its assigned congestion type, accumulated occurrences of congestion are determined. The methodology is applied to data derived from speed sensors along the Bavarian freeway A9 in Germany. The results show a high share of Stop-and-Go traffic in the Greater Munich Area. All over the considered stretch, Jam Waves occur frequently, limited to a few locations but widely spread in time.

Keywords: Data Mining and Data Analysis, Travel Information, Travel Guidance, and Travel Demand Management
Abstract: Intelligent transportation management requires not only statistical information on users' mobility patterns, but also knowledge of their selected transportation modes. The latter can be inferred from users' GPS records, as captured by smartphone or vehicle sensors. The recently demonstrated prevalence of deep neural networks in learning from data makes them a promising candidate for transportation mode identification. However, the massive geospatial data produced by GPS sensors are typically unlabeled. To address this problem, we propose an unsupervised learning approach for transportation mode identification. Specifically, we first pretrain a deep Convolutional AutoEncoder (CAE) using unlabeled fixed-size trajectory segments. Then, we attach a clustering layer to the CAE's embedding layer, the former maintaining cluster centroids as trainable weights. Finally, we retrain the composite clustering model, encouraging the encoder's learned representation of the input data to be clustering-friendly by striking a balance between the model's reconstruction and clustering losses. By further incorporating features computed over each segment, we achieve a clustering accuracy of 80.5% on the Geolife dataset without using any labels. To the best of our knowledge, this is the first work to leverage unsupervised deep learning for clustering of GPS trajectory data by transportation mode.

Keywords: Data Mining and Data Analysis, Modeling, Simulation, and Control of Pedestrians and Cyclists, Other Theories, Applications, and Technologies
Abstract: Human trajectory prediction is a quite challenging task mainly due to numerous social interactions and plausible paths in complex crowed scenarios and varying environments. Data-driven machine learning approaches based on Recurrent Neural Networks (RNNs) have, recently, achieved significant results in modelling human-human interactions in a scene. However, information-sharing pooling modules across RNN Encoders which operate on a local spatial context fail to model long-term scene level correlations, while other that adopt a more global approach are restricted to a rather simplistic formulations due to high computational costs. In this work, we introduce a novel pooling mechanism designed to perform trajectory pooling on a higher semantic level. We have developed a novel multi-layer network architecture based on a new Edge Convolutional operator acting on irregular data which is able to generalize local human-human interactions on a semantic social context. To assess the performance of the proposed social pooling with edge convolutions, we have integrated it into a state-of-the-art trajectory prediction framework based on Generative Adversarial Networks (GANs). Our module managed to overall outperform, by a significant margin, several state-of-the-art pooling modules in real-world challenging benchmark datasets.

Keywords: Intelligent Logistics, Off-line and Online Data Processing Techniques, Simulation and Modeling
Abstract: In this paper we address the problem of same-day pick-up and delivery where a set of tasks are known a priori and a set of tasks are revealed during operation. The vehicle routes are precomputed based on the known and predicted requests and adjusted online as new requests are revealed. We propose a novel anticipatory insertion method which incorporates a set of predicted requests to beneficially adjust the routes of a fleet of vehicles in real-time. Requests are predicted based on historical data, which is clustered in advance. We exploit inherent patterns of the demand, which are captured by historical data and include them in a dynamic vehicle routing solver based on heuristics and adaptive large neighborhood search. The proposed method is evaluated using numerical simulations on a variety of real-world problems with up to 1655 requests per day. Their degree of dynamism ranges from 0.70 to 0.93. These instances represent dynamic multi-depot pickup and delivery problems with time windows. The method has shown to require less driven kilometers than comparable methods.

Keywords: Intelligent Logistics
Abstract: On-demand mobile facility services are a promising approach to mitigate social problems related to transportation. Route optimization to satisfy customer demands is an essential technology to realize the services. Most studies of the route optimization for the services have been focused on finding a better assignment from vehicles to customers and a better order of visiting customer locations under the assumption that the customers waiting at the locations without moving. In this paper, we formulate cooperative routing problem between customers and vehicles, which minimizes total travel cost by optimizing both vehicle and customer routes. We also propose a heuristic approach to find solutions for large instances. We demonstrate that customer cooperation helps to reduce the total travel cost compared to a solution of standard vehicle routing problem in synthetic experiments using the road network of Manhattan, NY, USA. We confirmed that the total travel distance of the customers and the vehicles was reduced by 20% using our heuristics comparing to solutions of the vehicle routing problem with little extra computational cost.

Keywords: Intelligent Logistics, Human Factors in Intelligent Transportation Systems, ITS Policy, Design, Architecture and Standards
Abstract: For transport logistics, often the most inefficient part of the journey is the route between distribution centre and end customer. This route, referred to as last-mile delivery, generally uses smaller goods vehicles, to deliver low-volumes to multiple destinations. To optimise this process, route planning optimisation software is used, to maximise the number of deliveries a driver can complete in a day. To further optimise this process, companies are starting to test autonomous goods vehicles (AGVs). This paper presents an evaluation of the impact and barriers of AGVs for last-mile delivery in the UK, by conducting a study of people in the logistics industry and experts in autonomous technology. Qualitative analysis is used to identify positive and negative impacts of the introduction of driverless AGVs, and barriers, in terms of government policy and technical restrictions, which could slow down wide-scale adoption. From the results, we find logistics companies are being pressured to reduce lead-times and offer more predictable delivery-times. This is increasing pressure on the workforce, which already has high-turnover and difficulties in recruitment. Therefore, AGVs are considered a solution to a present problem, which is preventing logistics companies growing and achieving delivery targets, driven by public demand.

Keywords: Intelligent Logistics, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: As an important part of large-scale intelligent warehouse system, path planning by considering the cooperation among automated guided vehicles (AGVs) becomes an important factor to enhance the efficiency of the system. To this end, we propose a novel path planning scheme based on space-time map with the target of improving the path planning efficiency. Specifically, we first model the time dimension and construct a space-time map to obtain the planned path information of the intelligent warehouse system. Then, by taking the size of AGV and turning cost into consideration, we design a node extension algorithm to limit the search direction of AGVs. To decrease the complexity of the proposed algorithm and improve the efficiency of head-on conflict avoidance, a time window based piecewise path planning method and a mechanism of protected zone are developed, respectively. Simulation results show that the proposed space-time map based path planning scheme has a better performance than the conventional method in terms of the number of turns, the system running time and the moving distance of AGVs.

Keywords: Cooperative Techniques and Systems, Road Traffic Control, Intelligent Logistics
Abstract: Automated driving has started to be used commercially for individual mobility and public transport in recent years. As soon as automated driving is commercially exploited, automated vehicle fleets require assistance which can be provided by an operational control center as known from air traffic, public transport or process technology. The need for assistance is shown by two aspects. The aspects are human interaction and efficiency of journeys. Based on that, a control center is proposed to address those aspects. Further, a concept for a control center is derived. For that, the methods of interviewing experts, observation and literature research are used. Specific tasks for the control center are defined. These tasks were examined and categorized into three service categories. The three service categories are emergency service, fleet service and teleoperation service. Due to the categories, future-built control centers will be scalable and adaptable to its demand. Whereas two of three categories are well covered by industry or research, the teleoperation service as the essential problem solving technique needs further development. Moreover, further research will be required to quantify the control center demand.

Keywords: ITS Field Tests and Implementation, Sensing, Vision, and Perception, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: This paper presents the Autonomous Siemens Tram that was publicly demonstrated in Potsdam, Germany during the InnoTrans 2018 exhibition. The system was built on a Siemens Combino tram and used a multi-modal sensor suite to localize the vehicle, and to detect and respond to traffic signals and obstacles. An overview of the hardware and the developed localization, signal handling, and obstacle handling components is presented, along with a summary of their performance.

Keywords: ITS Field Tests and Implementation, Electric Vehicles, Personalized Public Transit
Abstract: Since mid-June 2019, electric scooters have been permitted on German roads. Many companies offer these in the form of sharing vehicles in Germany's major cities. These justify their existence through the zero-emission alternative to cars. How these vehicles are accepted, how they are used and whether they actually contribute to the transformation of the German traffic is being analysed in this paper. An online survey in Munich shows that e-scooters mainly replace walking and public transport and despite their large presence in the city landscape they are not used often. It appears that e-scooters are perceived as a leisure/fun object and less safe than bikes. The introduction of parking spaces with integrated charging facilities could save the bad reputation of electric scooters as deduced in the study. This reputation covers the environment, safety and the cityscape. In this study is also shown how the pricing model could be traced back to the absence of first mile last mile (FMLM) using.

Keywords: ITS Field Tests and Implementation, ITS Policy, Design, Architecture and Standards, Other Theories, Applications, and Technologies
Abstract: The important role of Information Technologies (IT) on the research and innovation (R&I) as well as on systems planning and implementation (SP&I) and operation of Intelligent Transportation Systems (ITS) is well known and almost by definition accepted. The relationship between IT and ITS, has been the subject of numerous research investigations and publications most of them relying on qualitative assessments and realizations about the nature and extent of the relationship. This paper is primarily concerned with the development and testing of a novel quantification methodology for the calculation (estimation) of the extent to which IT contributes to ITS (R&I and/or SP&I). The proposed methodology is based on the tools and concepts provided by the theory of production and more particularly on the calculation of the Productive Efficiency of the ITS sector within a given geographical area, with and without the contribution of IT. After a brief literature review, the theoretical formulation for the estimation procedure is given and then it is tested with data drawn from a wider questionnaire survey which the author has recently conducted in the frame of another of his research studies focused on innovation production in the ITS sector.

Keywords: ITS Field Tests and Implementation, Other Theories, Applications, and Technologies, Sensing, Vision, and Perception
Abstract: End-to-end autonomous driving approach, which directly maps raw input images to vehicle control commands using deep neural networks, is gaining considerable attention from both academia and industry. Researchers have conducted studies on this subject over the past few years. However, they have focused on designing network architecture, and evaluated performance only with root mean square error (RMSE), which did not account for the temporal dynamics of autonomous driving. In this paper, we propose a time-to-line crossing (TLC) enhanced end-to-end driving framework. The proposed framework is original to the industry in three ways. First, for the fine-labeled training dataset for end-to-end autonomous driving, the TLC based label correction algorithm is applied to reduce the inaccuracies of driver action, which is used as a ground truth. Second, we designed a novel deep neural network model based on bi-directional convolutional long short-term memory (bi-CLSTM) which can sufficiently encode the spatial and temporal features in the input image sequence. Third, in addition to the RMSE evaluation metric, we validated the performance of the end-to-end driving model using TLC, the advanced driving support system (ADAS), and a driver performance indicator. We integrated the proposed framework with existing end-to-end driving models on a full-scale autonomous vehicle in the experimental portion of our study. The results show that the proposed framework is valid and that our network model outperforms the existing models in terms of both RMSE and TLC.

Keywords: ITS Field Tests and Implementation, Sensing and Intervening, Detectors and Actuators, Off-line and Online Data Processing Techniques
Abstract: Performing high level autonomous navigation in a reliable and robust way considering different ambient conditions is a very challenging task. To achieve this goal, a mix of different sensors, such as cameras, lidars, and radars, are normally used to gather information from the environment. Since each sensor works based on different physical principles, they are affected differently by the challenging conditions, like weather interference for example. Looking to explore the influence of high intensity rain (98mm/h), this paper presents a robust experimental protocol that analyzes the influence inside the near field of lidar and radar sensors. The results shows how the effect of rain droplets degrades the backscattering signal from both sensors, affecting the information regarding the target's dimension. The consequences in terms object and feature detection´s changes are also discussed.

Keywords: ITS Field Tests and Implementation, Simulation and Modeling, ITS Policy, Design, Architecture and Standards
Abstract: Digital twins and parallel system have already been regarded as one of the most effective approaches that give a great impetus to the development of transportation system, especially for the testing of vehicle intelligence. State synchronization, as the main influencer of real-time interaction in a parallel system, determines the testing accuracy and computational efficiency. Despite the fact that the synchronization control is already a well-explored field, the traffic state synchronization in the application of the vehicle testing via virtual-real interaction is still a topic for further research. In this paper, to achieve better synchronization, a path generation method based on the Frenet frame is firstly designed to decompose the object's motion into the longitudinal and lateral directions, and achieves the trajectory tracking based on the near real-time data sent from the physical space. Then to eliminate the stochastic latency, the advance estimate-based path modification method is proposed to generate a stretch of path in advance. Finally, the integrated approach is implemented in a testing platform and the experimental results prove that the proposed method improves the synchronous rate by an average of 78.4% and 63.7% in the scenario of straight driving and lane changing.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Simulation and Modeling
Abstract: In the era of automation, autonomous point – to – point shuttles are said to be among first mobility on demand service that will emerge. But, what will be the impacts of such a service in the implementation area? The scope of the present paper is to assess the impacts of an autonomous shuttle bus service on traffic conditions, road safety and environment. For this purpose, a shuttle bus route was designed to operate in a part of the Athens road network and various scenarios have been developed including peak and off peak hours, existence of a dedicated lane for the shuttle bus, incident occurrence as well as different penetration rates and profiles of autonomous vehicles. Results indicate that the autonomous shuttle bus operation leads to increased delay times on its route. The speed variance of shuttle bus and the prevailing traffic vehicles is up to 25 km/h during off peak hour. It is also shown that if the shuttle bus uses a dedicated lane, both the delay time and CO2 emissions are decreased. Automation decreases CO2 emissions during peak hour conditions and improves road safety, as the number of conflicts is reduced.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Aerial, Marine and Surface Intelligent Vehicles, Network Modeling
Abstract: In this paper we present a decision support system for essential planning processes of innovative transportation systems. We enhance an existing transportation model to analyze how passenger drones and autonomous cars can interact optimally to enable mobility and reduce congestion. The resulting transportation model covers 70,550 square kilometers as well as 12.6 million citizens and extends passenger drone services beyond urban use cases. Building upon this customization, we first identify a set of potential intermodal vertihub locations for vertical take-off and landing (VTOL). Second, a mathematical model is developed to decide at which locations vertihubs should be built to enable access to the new mode for a maximum number of citizens. Third, we connect the vertihubs through a fine-meshed flight route network using Delaunay triangulation. We finally apply the model to perform specific analyses concerning the interactions of these future modes of transport: With optimized vertihub locations and 30 min autonomous feeder service, more than 70% of the population could have access to passenger drones. Moreover, we perform sensitivity analyses for feeder time parameters and a possible substitution of public transport (PT) by drones. Finally, we identify a master vertihub location with a minimal flight distance to all other vertihubs.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Sensing, Vision, and Perception, Driver Assistance Systems
Abstract: This paper presents a Real-Time Bird's Eye View Multi Object Tracking (MOT) system pipeline for an Autonomous Electric car, based on Fast Encoders for object detection and a combination of Hungarian algorithm and Bird's Eye View (BEV) Kalman Filter, respectively used for data association and state estimation. The system is able to analyze 360 degrees around the ego-vehicle as well as estimate the future trajectories of the environment objects, being the essential input for other layers of a self-driving architecture, such as the control or decision-making. First, our system pipeline is described, merging the concepts of online and real-time DATMO (Deteccion and Tracking of Multiple Objects), ROS (Robot Operating System) and Docker to enhance the integration of the proposed MOT system in fully-autonomous driving architectures. Second, the system pipeline is validated using the recently proposed KITTI-3DMOT evaluation tool that demonstrates the full strength of 3D localization and tracking of a MOT system. Finally, a comparison of our proposal with other state-of-the-art approaches is carried out in terms of performance by using the mainstream metrics used on MOT benchmarks and the recently proposed integral MOT metrics, evaluating the performance of the tracking system over all detection thresholds.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Simulation and Modeling, Travel Behavior Under ITS
Abstract: Interactions between road users are both highly non-linear and profoundly complex, and there is no reason to expect that interactions between autonomous vehicles will be any different. Given the recent rapid development of autonomous vehicle technologies, we need to understand how these interactions are likely to present themselves, and what their implications might be. This paper looks into the impact of autonomous vehicles with differing controllers, focusing specifically on the effects of changing the mean and heterogeneity of controller parameters on three key performance metrics: throughput, passenger safety and comfort. Towards this end, we develop a method for systematically sampling vehicle controllers as a function of parameter heterogeneity. In addition to evaluating the impact of heterogeneity on performance, we quantify the relative impacts of controller input parameters on the output performance metrics by means of sensitivity analyses. The MovSim traffic simulator was used to simulate a realistic traffic system, whilst recording maximum throughput, as well as lane change frequencies and mean absolute accelerations as proxies for safety and comfort. Our results reveal that traffic performance is primarily affected by the heterogeneity of vehicle target velocities, as well as by the mean values of a very small subset of the parameters, of which the target velocity was by far the most significant.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Cooperative Techniques and Systems, Simulation and Modeling
Abstract: One way to improve traffic conditions is to encourage drivers to join platoons, i.e. strings of connected vehicles (CVs) maintaining a short headway on the road. In this paper, we introduce an optimization algorithm that assigns CVs that wish to join a platoon to existing nearby platoons such that the total travel time of all traffic participants is optimized. We assume that platoons are travelling on a dedicated high-occupancy platoon (HOP) lane and are capable of vehicle-to-infrastructure (V2I) communication. Our algorithm consists of two layers, where the first optimizes for which platoon and its side (front/back) a CV could join, and the second decides which of the participating CVs should merge based on a emph{score} function (estimated merging time). Simulations show that our algorithm achieves up to a 20% travel time reduction compared to a random merging procedure and that 70% of time reduction comes from an optimized vehicle selection (second layer).

Keywords: Sensing, Vision, and Perception, Advanced Vehicle Safety Systems, Driver Assistance Systems
Abstract: This work shows and analyzes the LiDAR performance in real-world heavy winter conditions captured in Northern Europe. We review how low temperatures, salted roads and turbulent snow in front of a passenger car influence LiDAR systems developed for automated driving functions. Two test cars were driven in the north of Finland and Sweden for 1.5 weeks to gather a large amount of point cloud data in different urban and rural scenarios. We show that the benchmarked LiDAR sensors have surprising performance differences in winter. Some of the sensors got mechanically frozen whereas others went out of the measurement range and were completely blind. Especially the latest multi-layer sensors showed significant problems. We propose countermeasures such as heating and protecting in order to improve the performance and suggest how the software can take the performance degradation into account.

Keywords: Sensing, Vision, and Perception, Accurate Global Positioning, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: In this paper we explore how visual simultaneous localisation and mapping (vSLAM) systems might estimate the pose (position and orientation) of a vehicle surveying highway assets, as part of a wider transport asset management (TAM) system. Such a system would reduce reliance on (or enhance estimation from) a GPS-enabled inertial measurement unit (IMU). Two problems with employing vSLAM systems on highway survey imagery are identified. Firstly, straight segments of the highway cause low parallax issues, and secondly the presence of other vehicles breaks the assumption that the camera moves within a static environment. To overcome these problems, an existing monocular vSLAM system is modified with two state-of-the-art deep neural network (DNN) tools. We show that the modified system provides an improved estimation of pose. In addition to TAM, our work has clear applications in autonomous vehicles, when there is only limited visual field sensing.

Keywords: Sensing, Vision, and Perception
Abstract: Recently, convolutional neural network (CNN) based methods have achieved superior results in generic object detection and have become the de-facto standard in the domain. However, potential adaptations to industrial areas are not well studied yet. A case worth exploring is the train component detection, in which the components may have strong relationships and some components (e.g., screws and nuts) are very small. Nevertheless, the detection performance of small train components significantly affects the efficiency of overall train component detection. In this work, we propose a novel robust train component detection(RTCD) framework, built on cascading CNNs and utilizing prior structure knowledge of the relationships between train components. The core idea of RTCD is to detect the big and easily detectable component first, and then find the areas that may contain small and challenging to detect components for following fine-grained exploitation. Our proposed attention region mechanism can find regions deserving of further analysis based on the region-of-interest (ROI) detected by the previous CNNs with the known structure knowledge. Then, these areas are cropped, zoomed in and fed into the following deep learning models for further detection. In order to verify the effectiveness of RTCD, 1,130 high-resolution images of moving trains are captured and collected, from which 17,334 critical train components are manually annotated. Extensive experiments therein have demonstrated that RTCD outperforms the existing state-of-the-art baselines significantly. The dataset and corresponding source code will be released to facilitate more future work.

Keywords: Sensing, Vision, and Perception, Driver Assistance Systems, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: LiDARs are usually more accurate than cameras in distance measuring. Hence, there is strong interest to apply LiDARs in autonomous driving. Different existing approaches process the rich 3D point clouds for object detection, tracking and recognition. These methods generally require two initial steps: (1) filter points on the ground plane and (2) cluster non-ground points into objects. This paper proposes a field-tested fast 3D point cloud segmentation method for these two steps. Our specially designed algorithms allow instantly process raw LiDAR data packets, which significantly reduce the processing delay. In our tests on Velodyne UltraPuck, a 32 layers spinning LiDAR, the processing delay of clustering all the 360^circ LiDAR measures is less than 1ms. Meanwhile, a coarse-to-fine scheme is applied to ensure the clustering quality. Our field experiments in public roads have shown that the proposed method significantly improves the speed of 3D point cloud clustering whilst maintains good accuracy.

Keywords: Sensing, Vision, and Perception, Driver Assistance Systems, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: In this paper, we propose a novel radar-based lane estimation method using Deep Neural Network(DNN) without vision sensors. First, the feature vector is selected using data coming from radar and in-vehicle sensors. The feature vectors are stacked and entered into the network so that the input of the network has spatio-temporal information of the relative motion between the ego vehicle and a leading vehicle. We used a parallel structure of the DNN to estimate the road lane model for the Lane-Keeping System(LKS). The Scaled Conjugate Gradient method is adopted for optimizing the neural network. We performed a comparative study between a vision sensor and the proposed system. From the experiment results, the proposed scheme outperforms the vision system when the vision system becomes failure due to environmental effects such as shadows or lane contamination. It is expected that the proposed method is sufficient to improve the performance of LKS if the proposed system is fused with the vision system for fail-operational lane-keeping system of autonomous highway driving.

Keywords: Simulation and Modeling, Data Mining and Data Analysis, Network Modeling
Abstract: Short-term traffic forecasting is one of the key functions in Intelligent Transportation System (ITS). Recently, deep learning is drawing more attention in this field. However, how to develop a deep learning based traffic forecasting model that can dynamically extract explainable spatial correlations from traffic data is still a challenging issue. The difficulty mainly comes from the inconsistency between static model structures and the dynamic evolution of traffic conditions. To overcome this difficulty, we proposed a novel multistep speed forecasting model, Dynamic Graph Filters Networks (DGFN). The major contribution is that the regular pixel-wise dynamic convolution is extended to graph topology. DGFN has a simple recurrent cell structure where local area-wide graph convolutional kernels are dynamically computed from varying inputs. Experiments on ring freeways show that DGFN is able to precisely predict short-term evolution of traffic speed. Furthermore, we theoretically explain why DGFN is not a pure "black-box", but a "gray-box" model that actually reduces entangled spatial and temporal features into one component representing dynamic spatial correlations. It permits tracking real-time interactions among adjacent links. DGFN has the potential to relate trained parameters in deep learning models with physical traffic variables.

Keywords: Simulation and Modeling, Network Modeling
Abstract: This paper presents OTM-MPI, an extension of the Open Traffic Models platform (OTM) for running macroscopic traffic simulations in high-performance computing environments. OTM-MPI represents the first open-source, distributed-memory, macroscopic simulation model developed for modern high performance parallel machines and large networks. Macroscopic simulations are appropriate for studying regional traffic scenarios when aggregate trends are of interest, rather than individual vehicle traces. They are also appropriate for studying the routing behavior of classes of vehicles, such as app-informed vehicles. The network partitioning was performed with METIS. Inter-process communication was done with MPI (message-passing interface). Results are provided for two networks: one realistic network which was obtained from Open Street Maps for Chattanooga, TN, and another larger synthetic grid network. The software recorded a speedups of 198x using 256 cores for Chattanooga, and 475x with 1,024 cores for the synthetic network.

Keywords: Simulation and Modeling, ITS Field Tests and Implementation, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: With the widespread of research in the field of autonomous vehicles the value and impact of various simulators increase dramatically as they allow for quick and safe experimentation with the design of a vehicle, environment and driving scenarios. In this paper, the authors demonstrate how autonomous vehicle research and development can be facilitated by open-source robot simulators based on the experience gained from a robotized KAMAZ NEO truck industrial project. In particular, the Webots robot simulator was applied for 3D reconstruction of the experimental test-site for vehicle motion simulation and development of a web-based dashboard for controlling and monitoring the autonomous vehicle both in the simulation and the real-world.

Keywords: Simulation and Modeling, Data Mining and Data Analysis, Theory and Models for Optimization and Control
Abstract: Shared Electric Connected and Automated Vehicles have the potential to improve transportation safety, mobility, and energy efficiency. A plug-in hybrid electric architecture is well suited for developing connected and automated vehicle (CAV) applications, allowing for vehicle dynamics management and powertrain control. In this paper, we developed a data-driven modularized modeling approach for a plug-in hybrid electric bus (PHEB), thereby allowing for a wide range of connected and automated vehicle applications. Instead of using an end-to-end learning approach to model the PHEB, our modularized modeling approach considers the physical connection of each component of PHEB, which provides various signals and dynamics of each subsystem for testing use or controller design. The plug-and-play (PnP) feature allows us to customize the bus model and update each individual module in a flexible manner. The modules include human driver behavior, energy management system, internal combustion engine, electric motor(s), transmission, and powertrain dynamics. For each module, a Long Short-term Memory (LSTM) network is utilized to learn each modules’ behavior and dynamics using the data from extensive dynamometer-in-the-loop (DiL) testing.

Keywords: Simulation and Modeling
Abstract: The importance of simulation-based approaches for the development of automated driving functions has strongly increased in recent years. Since current validation methods probably will not be feasible for highly automated driving functions, the simulation-based techniques' relevance for the validation procedure is also increasing. In this context, it is a large challenge to simulate the huge if not infinitely large parameter space of an automated vehicle's (AV) surrounding. Therefore, this paper presents a model which allows the modeling of the statistics of traffic scenes on straight highway sections. The model is based upon recent approaches using Bayesian networks and is adapted to the usage of sum-product networks (SPNs). It is shown, that the use of SPNs can enhance the sampling speed up to 36 times compared to the Bayesian network baseline.

Keywords: Theory and Models for Optimization and Control, Road Traffic Control, Sensing and Intervening, Detectors and Actuators
Abstract: In this paper we present a numerical study on control and observation of traffic flow using Lagrangian measurements and actuators. We investigate the effect of some basic control and observation schemes using probe and actuated vehicles within the flow. The aim is to show the effect of the state reconstruction on the efficiency of the control, compared to the case using full information about the traffic. The effectiveness of the proposed state reconstruction and control algorithms is demonstrated in simulations. They show that control using the reconstructed state approaches the full-information control when the gap between the connected vehicles is not too large, reducing the delay by more than 60% when the gap between the sensor vehicles is 1:25 km on average, compared to a delay reduction of almost 80% in the full-information control case. Moreover, we propose a simple scheme for selecting which vehicles to use as sensors, in order to reduce the communication burden. Numerical simulations demonstrate that with this triggering mechanism, the delay is reduced by around 65%, compared to a reduction of 72% if all connected vehicles are communicating at all times.

Keywords: Theory and Models for Optimization and Control, Off-line and Online Data Processing Techniques, Data Mining and Data Analysis
Abstract: A key problem in short-term traffic prediction is the prevailing data missing scenarios across the entire traffic network. To address this challenge, a transfer learning framework is currently used in the literature, which could improve the prediction accuracy on the target link that suffers severe data missing problems by using information from source links with sufficient historical data. However, one of the limitations in these transfer-learning based models is their high dependency on the consistency between datasets and the complex data selection process, which brings heavy computation burden and human efforts. In this paper, we propose an adaptive transfer learning method in short-term traffic flow prediction model to alleviate the complex data selection process. Specifically, a self-adaptive neural network with a novel domain adaptation loss is developed. The domain adaptation loss is able to calculate the distance between the source data and the corresponding target data in each training batch, which can help the network to adaptively filter inconsistent source data and learn target link related information in each training batch. The Maximum Mean Discrepancy (MMD) measurement, which has been fully validated and applied in transfer learning research, is used in combination with the Gaussian kernel to measure the distance between datasets in each training batch. A series of experiments are designed and conducted using 15-minute interval traffic flow data from the Highways England, UK. The results have demonstrated that the proposed adaptive transfer learning method is less affected by the inconsistency between datasets and provides more accurate short-term traffic flow prediction.

Keywords: Theory and Models for Optimization and Control, Automated Vehicle Operation, Motion Planning, Navigation, Simulation and Modeling
Abstract: This paper investigates how to compensate for curvature response mismatch in lateral Model Predictive Control (MPC) of an autonomous vehicle. The standard kinematic bicycle model does not describe accurately the vehicle yaw-rate dynamics, leading to inaccurate motion prediction when used in MPC. Therefore, the standard model is extended with a nonlinear function that maps the curvature response of the vehicle to a given request. Experimental data shows that a two Gaussian functions approximation gives an accurate description of this mapping. Both simulation and experimental results show that the corresponding modified model significantly improves the control performance when using Reference Aware MPC for autonomous driving of a Scania heavy-duty construction truck.

Keywords: Theory and Models for Optimization and Control, Simulation and Modeling, Travel Behavior Under ITS
Abstract: This study developed and compared eco-driving strategies for different traffic-management measures. Results show that eco-driving in a dedicated lane can substantially reduce energy consumption, which can be improved further by providing the vehicle with information regarding the phase-timing of upcoming traffic lights. For vehicles operating in mixed traffic, the energy savings strongly depend on the interaction with other traffic participants. Results show that an eco-driving strategy that limits the maximum inter-vehicle distance leaves less opportunity for eco-driving, and barely benefits from traffic light information. An eco-driving strategy without a maximum-inter-vehicle distance results in higher energy savings and does benefit from traffic light information, but leads to large inter-vehicle distances, which may induce congestion. Generating detailed results on the impact of eco-driving in traffic requires implementing the algorithms in agent-based traffic simulations.

Keywords: Theory and Models for Optimization and Control, Automated Vehicle Operation, Motion Planning, Navigation, Other Theories, Applications, and Technologies
Abstract: Control Barrier Functions (CBF) have been recently utilized in the design of provably safe feedback control laws for nonlinear systems. These feedback control methods typically compute the next control input by solving an online Quadratic Program (QP). Solving QPs in real-time can be a computationally expensive process for resource-constrained systems. In the presence of disturbances, finding CBF-based safe control inputs can get even more time consuming as finding the worst-case of the disturbance requires solving a nonlinear program in general. In this work, we propose to use imitation learning to learn Neural Network based feedback controllers which will satisfy the CBF constraints. In the process, we also develop a new class of High Order CBF for systems under external disturbances. We demonstrate the framework on a unicycle model subject to external disturbances, e.g., wind or currents.

Keywords: ITS Policy, Design, Architecture and Standards, Other Theories, Applications, and Technologies
Abstract: Intelligent Transport Systems (ITS) and Cooperative Intelligent Transport Systems (C-ITS) are of high significance, mainly due to the benefits they have in terms of operation of the transport network. Despite ITS and C-ITS importance in the operation of the transport network, there is a major knowledge gap regarding their development, way of operation and significance worldwide and especially among the responsible for their deployment public authorities. In order for such fragmentations to be tackled, an online training platform concerning the operation and impacts of ITS and C-ITS has been designed in the framework of the European Union’s Horizon 2020 funded CAPITAL project. For the proper design of the CAPITAL Online Training Platform, two main approaches have been studied: capacity building and massive open online courses. The present study provides insight regarding the design and the context of the CAPITAL Online Training Platform.

Keywords: ITS Policy, Design, Architecture and Standards
Abstract: The rising complexity nested within the contemporary transport sector, due to the increasing number of stakeholders involved, has led to an increase in the amount of generated data. This constitutes an opportunity for finding new ways of utilizing previously wasted data. A perfect fit may be found in Intelligent Transport Systems (ITS), a multidisciplinary and expanding area of research. ITS produce a number of benefits; however, such systems require continuous and reliable data streams. The challenge, therefore, lies in identifying underutilized data sources and creating efficient, consistent, and secure connections between data providers and consumers. This need has been recognized by the European Union, which suggests the creation of National Access Points (NAPs) for the distribution of ITS data. This paper aims to introduce the concept of a NAP and to review and analyze the various NAPs of European Union Member States in order to gain insight concerning their level of deployment. To this end, a series of proposed variables and their associated values were recorded for each Member State. Results show a trend for an increased rate of data exchange. However, they also reveal a heterogeneous design philosophy among the various platforms.

Keywords: ITS Policy, Design, Architecture and Standards, Simulation and Modeling
Abstract: In modern cars, software functions and services account for a large part of value creation and competitive differentiation. Several tools exist to address the development of such electrical/electronic architectures (E/E-A). In industry, the proprietary tool PREEvision developed by Vector Informatik GmbH is widely used to support the development for AUTOSAR, while in science and research, tools and ecosystems such as the Robot Operating System (ROS) are preferred because of their open-source nature. This leads to a multitude of freely available ROS components whose reusability in industrial AUTOSAR-based projects is desirable. Therefore, in this paper we present an approach to transform models between both worlds and thus to link them. This enables the further use of already existing components.

Keywords: ITS Policy, Design, Architecture and Standards
Abstract: Despite the debate regarding the timeframe and rate of penetration of Autonomous Vehicles, their potential benefits and implications have been widely recognized. Therefore, assessing the readiness of individual countries to adopt such technologies and adapt to their introduction is of particular importance. This paper aims to enrich our understanding of EU readiness regarding the introduction of autonomous vehicle technologies by assessing the case of Greece. Thus, through a literature review, the criteria upon which such an assessment should be based are established and analysed. Subsequently, the case of Greece is assessed based on those criteria by finding relevant sources that support and justify any assessment. Regardless of the outcome concerning the readiness of Greece, such an assessment should help identify areas in which focus should be given in order to ensure a smoother transition to such technologies. This contribution is expected to assist policy makers worldwide.

Keywords: Transportation Security
Abstract: Cooperative intelligent transportation system (C-ITS) applications are generally susceptible to position spoofing-dependent attacks such as Sybil and DDoS attacks due to a lack of established solutions. This paper presents a novel cyber-physical blockchain cryptographic architecture to help prevent position spoofing attackers from becoming validated nodes in C-ITS applications. The solution also guarantees security requirements including the non-trivial non-repudiation in light of these and other attacks. With a use case of electronic toll collection (ETC), our architecture implements techniques based on Received Signal Strength Indication (RSSI) measurements in conjunction with blockchain authentication methods such as Proof-of-Location and smart contracts to determine the legitimacy of a node. We demonstrate our solution in experiments using ITS-G5 Cohda Wireless technology (a Road Side Unit and two On-Board Units programmed with the ITS Vanetza stack) with functionalities specified by the European Telecommunications Standardization Institute (ETSI). From our experimental results from several driving-based data gathering tests, we discovered that our solution is able to cope with noise and relative velocity challenges because it incorporates both OBUs and RSUs in the Proof of Location computation steps. In light of this, the proposed architecture may also be applicable to govern V2X in general.

Keywords: Modeling, Simulation, and Control of Pedestrians and Cyclists, Commercial Fleet Management, Personalized Public Transit
Abstract: This paper explores the feasibility of a free-floating electric bicycle sharing system. They key enabling factor is to reduce the costs associated to recharging the batteries by applying the charge sustaining paradigm. The paper derives a model that describes the energy dynamics of a fleet of 300 bicycles. The model first guides the design of an Energy Management System and then quantitatively analyzes the feasibility of the bike-sharing system. In particular, we show that, although it is impossible to completely remove maintenance interventions, a rental rate of 2.5 picks per bicycle per day is enough to guarantee charge sustaining at the fleet level.

Keywords: Modeling, Simulation, and Control of Pedestrians and Cyclists, Simulation and Modeling, Theory and Models for Optimization and Control
Abstract: The Personal Mobility Vehicles (PMV) and in particular the electric scooters enjoy increasing popularity and their use has become widespread in the urban environment. The use of existing infrastructure, such as the sidewalks, by e-scooter drivers, poses a new challenge to policy makers trying to regulate the use of this new mode of transport so that it will be smoothly integrated in the urban networks. So far, there is limited research on the movement of electric scooters and their interaction with pedestrians, depriving the authorities of tools to draw and enforce effective policies. In this paper, we explore the applicability of the social force model for pedestrian dynamics to simulate the movement of e-scooters and the interaction between e-scooters and pedestrians . To conduct this study, we extract electric scooter and pedestrian trajectories through image analysis of videos containing pedestrian and e-scooter movement. Based on the extracted trajectories, scenarios and the respective initial conditions are generated. The social force model is used for the scenarios, and simulated trajectories of e-scooter and pedestrian movement are produced. The simulated trajectories are compared to the experimental trajectories with the Root Mean Squared Error (RMSE). Finally, the parameters of the social force model and the free speed of the vehicle are estimated with the Cross Entropy Method (CEM).

Keywords: Modeling, Simulation, and Control of Pedestrians and Cyclists, Sensing, Vision, and Perception, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: In this work, we propose a constrained moving horizon state estimation approach to estimate a pedestrian's states in 3D with respect to a global stationary frame including position, velocity, and acceleration that are robust to intermittently noisy or absent sensor measurements. Utilizing a computationally light-weight fusion of a Deep Neural Network based 2D pedestrian detection algorithm and projected LIDAR depth measurements, the approach produces the required measurements relative to the vehicle frame and combines them with the rotation and translation information obtained via odometry. The performance of the proposed approach is experimentally verified on our dataset featuring urban pedestrian crossings, with and without ego vehicle motion.

Keywords: Modeling, Simulation, and Control of Pedestrians and Cyclists, Intelligent Logistics, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: Abstract— Connected and autonomous vehicles (CAV) is a key technology for this century. One of the main challenges is to define a smart interaction behavior of CAV with the other road users. The challenge is mainly raised at conflicting points where path of CAV intersects with the other users. Recent studies show interaction with humans is a big challenge. It not only requires a collision avoidance system but also more communicative behaviors of the CAV. More precisely, pedestrian needs to understand the intention of the incoming CAV whether it will cross first or not according to its speed profile. One way to overcome this issue is to design optimal trajectory control of the CAV that matches with the pedestrian expectation. However, such a design faces the non-linearity of the space sharing constraint. This paper uses sequence modeling based on Petri Net in order to overcome the problem. It also uses Hamiltonian analysis to derive the optimal control. Numerical examples of the communicative behavior of CAV at pedestrian crossing show that the proposed approach provides CAV with a kind of automatic courtesy.

Keywords: Modeling, Simulation, and Control of Pedestrians and Cyclists, Human Factors in Intelligent Transportation Systems, Travel Behavior Under ITS
Abstract: Trajectory prediction in urban mixed-traffic zones (a.k.a. shared spaces) is critical for many intelligent transportation systems, such as intent detection for autonomous driving. However, there are many challenges to predict the trajectories of heterogeneous road agents (pedestrians, cyclists and vehicles) at a microscopical level. For example, an agent might be able to choose multiple plausible paths in complex interactions with other agents in varying environments. To this end, we propose an approach named Multi-Context Encoder Network (MCENET) that is trained by encoding both past and future scene context, interaction context and motion information to capture the patterns and variations of the future trajectories using a set of stochastic latent variables. In inference time, we combine the past context and motion information of the target agent with samplings of the latent variables to predict multiple realistic trajectories in the future. Through experiments on several datasets of varying scenes, our method outperforms some of the recent state-of-the-art methods for mixed traffic trajectory prediction by a large margin and more robust in a very challenging environment. The impact of each context is justified via ablation studies.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems
Abstract: In order to predict the behavior of human drivers accurately, the autonomous vehicle should be able to understand the reasoning and decision process of motion generation of human drivers. However, most of the conventional prediction methods overlook this and focus on improving the prediction results using the given data, the historical information. In contrast, human drivers not only depend on the historical motion but also consider future predictions when handling interactions with other vehicles. In this paper, we propose a novel recursive RNN encoder-decoder prediction model that takes the initial future prediction results as inputs of second prediction computation. This feedback mechanism can be interpreted as a network sharing, which allows the model to refine or correct the predicted results iteratively. We use two encoders to analyze both of the historical information and future information, and the attention mechanism is employed to interpret interaction. Our experimental results with the NGSIM dataset demonstrate the recursive structure enhances prediction results effectively compare to the baselines based on the ablation study and state-of-the-art methods. Furthermore, we observe that the results improve successively as the model iterates.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Rail Traffic Management, Theory and Models for Optimization and Control
Abstract: This work proposes a model for exploiting some possibilities that can derive from the concept of virtual coupling in railways, in which trains can be dynamically grouped to form single convoys. In particular, in the considered scenario, convoys running on a railway line are formed by virtually coupling train carriages, and the capacity of each convoy can change dynamically by leaving or adding carriages at each station. In the paper, an original optimization model based on a mathematical programming formulation is proposed for a passenger railway line, with the aim of determining the optimal time-variant capacity of trains that satisfies the transport demand and avoids trains with unnecessary capacity. More in detail, the proposed system determines, for each link of the line, the optimal number of carriages that is needed, allowing maneuvers like the parking of unused carriages and the exchange of them among the trains. The paper presents also an example with the results derived from the implementation of the model for a simple case, done with the aim of testing and validating the model.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Theory and Models for Optimization and Control, Electric Vehicles
Abstract: Increasing attention to autonomous passenger vehicles has also attracted interest in an autonomous racing series. Because of this, platforms such as Roborace and the Indy Autonomous Challenge are currently evolving. Electric racecars face the challenge of a limited amount of stored energy within their batteries. Furthermore, the thermodynamical influence of an all-electric powertrain on the race performance is crucial. Severe damage can occur to the powertrain components when thermally overstressed. In this work we present a race-time minimal control strategy deduced from an Optimal Control Problem (OCP) that is transcribed into a Nonlinear Program (NLP). Its optimization variables stem from the driving dynamics as well as from a thermodynamical description of the electric powertrain. We deduce the necessary first-order Ordinary Differential Equations (ODE)s and form simplified loss models for the implementation within the numerical optimization. The significant influence of the powertrain behavior on the race strategy is shown.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems, Driver Assistance Systems
Abstract: In this paper, we propose a waypoint tracking scheme with a collision-avoidance system using the artificial potential field (APF) with speed planning. It is shown that there is little literature about the combination work of waypoint tracking and APF-based local path planner. We present a new path planning method to smoothly join the local path predicted by APF to avoid obstacles and the path given by irregular waypoints to eliminate severe yawing of the heading angle of the vehicle. The proposed speed planning combined with APF effectively prevents possible excessive tire slip angle of the vehicle driving at high speed. The simulation results show that the proposed path planning algorithm with speed planning is effective in collision avoidance with static and/or dynamic obstacles for a waypoint tracking scenario. This proposed method is validated with the vehicle dynamic lateral motion model in MATLAB/Simulink and CarSim.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Driver Assistance Systems, Network Modeling
Abstract: In this paper we introduce the concept of routing and pathfinding, which is used by the Lanelet2 map framework for automated driving.

While in conventional map frameworks the routing graph is explicitly predefined, Lanelet2 interprets it at runtime. This offers many advantages for automated driving, because it allows the map information to be viewed from different perspectives, such as other road users or for other traffic situations.

We also analyze what information is typically required by routing applications in automated driving, how this affects the shape of the routing graph, and how this information can be extracted efficiently.

The paper concludes with an overview of developments in high-definition maps. In our opinion, these are lagging behind progress in the other areas of automated driving. We hope that Lanelet2 continues to gain increasing acceptance in the community and thus helps to remedy this shortcoming.

Keywords: Data Mining and Data Analysis, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Sensing, Vision, and Perception
Abstract: Recent advances in machine learning research could significantly alter the railroad industry by deploying fully autonomous trains. To achieve effective interaction between self-driving trains and the environment, an accurate long-range railway detection should be provided. In this paper, we propose a framework for the rail tracks segmentation on high-resolution images (2168x4096). The announced approach accelerates inference speed 6 times, by using two neural networks. The proposed architecture and its training approach provide a long-range railway segmentation within 150 meters, achieving 20 fps. Also, we propose an auxiliary algorithm detecting possible paths among all the found ones. To determine which data labeling approach has a higher impact, additional experiments were performed. The proposed framework provides a balanced tradeoff between computing efficiency and performance in the railroad segmentation problem.

Keywords: Data Mining and Data Analysis
Abstract: Although trains are more efficient and convenient than other transportation, delays often occur. Accurately predicting the delay time of trains is of great significance to both dispatchers and passengers. The method for predicting the arrival delay time of trains is based on feature selection algorithm and machine learning. First, we collect train delay cases to sort out the delay factors. In addition to internal factors, external factors such as weather and signal failure are also considered. Then, an improved max-relevance and min-redundancy method (mRMR) is used for feature selection. Finally, we apply the method of weighted random forest (wRF) to predict the delay time. The results demonstrate that the feature selection algorithm has a prominent effect on improving the accuracy of the model, and the mean square error based on the weighted random forest has an improvement potential in forecast precision.

Keywords: Data Mining and Data Analysis, Network Management, Network Modeling
Abstract: Traffic flow prediction, particularly in areas that experience highly dynamic flows such as motorways, is a major issue faced in traffic management. Due to increasingly large volumes of data being generated every minute, deep learning methods have been used extensively in the latest years for both short and long term traffic flow prediction. However, such models, despite their efficiency, need large amounts of historical information to be provided, and they take a considerable amount of time and computing resources to train, validate and test. This paper presents two new spatial-temporal approaches for building accurate short-term predictions along a popular motorway in Sydney Australia, by making use of the graph structure of the motorway network (including exits and entries). Our proposed methods are proximity-based, and they use the most recent available traffic flow information of the upstream counting stations closest to a given target station. Where such information is not available they employ daily historical means instead. We show that for short-term predictions (less than 10 minutes into the future), our proposed graph-based approaches outperform state-of-the-art deep learning models, such as long-term short memory, convolutional neuronal networks or hybrid models.

Keywords: Data Mining and Data Analysis, Transportation Security, Human Factors in Intelligent Transportation Systems
Abstract: The automatic recognition of different riding patterns in the context of naturalistic riding studies (NRSs) facilitates the behavioral analysis of powered two-wheelers (PTW), which is a challenging problem. In the NRS context, various multivariate time series data are provided using an inertial measurement unit (IMU). Modeling the temporal dependency between riding patterns using state-of-the-art machine learning methods is not a straightforward task and requires the extraction of relevant features. In this article, we suggest the use of recurrent neural networks (RNNs) for modeling the temporal dependence between successive patterns without requiring manual feature engineering. Experiments are carried out using a real-world dataset of instrumented motorbikes. The analysis of the network activations and estimated weights allows us to describe the complex riding patterns. Furthermore, comparisons with state-of-the-art machine learning methods show the effectiveness of RNNs in the identification of riding patterns.

Keywords: Data Mining and Data Analysis, Off-line and Online Data Processing Techniques, Theory and Models for Optimization and Control
Abstract: Robust and accurate short-term traffic flow prediction plays an important role in ITS. In terms of promoting the prediction accuracy and stability, prediction models and traffic flow characteristics are of equal importance. However, most of the existing literature concentrate on the performance of models and ignore the predictability of traffic flow data itself. In this paper, in order to make a breakthrough in predicting traffic flow with large fluctuation, a traffic flow prediction model based on decomposition and reconstruction is established. First, empirical mode decomposition (EMD) and wavelet threshold (WT) methods are combined to produce an EMDW method, which can decompose and reconstruct original traffic flow into stable sub-sequences with enhanced predictability and reduced noise. Second, combining the proposed EMDW method with least square support vector machine (LSSVM), a hybrid EMDW-LSSVM model is introduced to conduct short term traffic prediction. At last, the field data of Beijing 2nd ring road are employed to conduct experiments, proving that the proposed decomposition and reconstruction method can dramatically improve the accuracy of short-term traffic flow prediction.

Keywords: Multi-modal ITS, ITS Field Tests and Implementation, Human Factors in Intelligent Transportation Systems
Abstract: Safety al Level crossing (LC) is a minor issue for the road sector since it represents less than 1% of the accident mortality, but it is highly important for the railway sector, which accounts for thousands of accidents and collisions every year. In total, more than 500 causalities occur every year in the surroundings of LCs in the United States and the European Union Member States combined. This paper presents a multimodal cooperative safety system to alert about the vicinity of a LC and, if any, an approaching train. The system processes spatial data of trains and private nearby LCs and generate alerts about the presence of a nearby LC and the estimated time of arrival for approaching trains. The system was tested in the LCs of Thessaloniki by professional taxi drivers during a 12-month period. Qualitative analyses indicate positive acceptance by the drivers as well strong perception of reliability and safety impact of the system.

Keywords: Multi-modal ITS, Simulation and Modeling, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: In recent years, significant progress in the development of connected and automated vehicles (CAVs) has been observed. In combination with new control strategies at intersection zones, CAVs have the potential to increase both safety and efficiency in urban traffic: Numerous studies that present and evaluate automated intersection management (AIM) show a significant reduction of vehicle delays. However, most of these studies assume a fully connected traffic environment and do not consider non-connected road users such as human-driven vehicles, pedestrians, and bicyclists. In this paper, novel strategies for integrating bicyclists into AIM are introduced and compared to two different fully actuated traffic signal controls. The presented strategies consist of a first-come, first-served strategy for vehicles in combination with traffic signals for bicyclists. Depending on the available information about approaching bicyclists, bicycle operations are included into the control in fixed cycles or on demand. A four-leg intersection is simulated using a microsimulation platform where all control strategies are implemented and tested. A set of scenarios considering different levels of vehicle and bicycle demand is evaluated. Results show that bicyclists can be included into AIM while guaranteeing a maximum bicyclist waiting time, and the level of service for vehicles can be significantly improved.

Keywords: Multi-modal ITS
Abstract: Vehicle-to-everything (V2X) technology is a promising technology for the automobile industry in this decade. V2X includes vehicle-to-vehicle (V2V), vehicle-to-network (V2N), and vehicle-to-infrastructure (V2I) modes of communication. However, relying on a single radio or V2X mode of communication is not desirable in the design of V2X systems of the future. In the absence of dedicated ITS communication infrastructure on the road side, one can rely on infrastructure-less technologies like Wi-Fi Direct for V2V communication and cellular technologies like 4G/5G for V2N communication. If direct V2V communication is not possible with other vehicles present in the vicinity, vehicles can rely on cellular technologies for indirect mode of communication (V2N). To reap in the benefits of both V2V and V2N, we propose a cooperative V2X architecture by opportunistically using either or both of cellular and 802.11 based radios present in the vehicles for reliable and scalable communication among the vehicles. The efficacy of the proposed V2X system is demonstrated by developing a collision warning solution named SafeNav. Simulation results suggest that it is worth using both V2V and V2N modes of communication to significantly improve the reliability of safety message exchange especially in hybrid scenarios i.e., vehicles having either or both of cellular (4G) and IEEE 802.11p based radios. An Android app has been developed to further demonstrate the use of the proposed system for collision avoidance in real-world settings.

Keywords: Multi-modal ITS, Public Transportation Management
Abstract: In order to increase the attractiveness and the resilience of multimodal transport systems, each transport mode should be exploited according to its peculiar characteristics in an integrated way with the other modes. In this regard, it is possible to imagine a “synchromodal framework” relying on a “rail backbone”, where the synchronization of the different modes is driven by rail transport. On the other hand, the limited number of path alternatives that characterizes the rail network makes it vulnerable to service disruptions and may result in a significant Level of Service loss. In this context, this paper proposes a methodology for urban public service providers that allows to design a multimodal service able to react, in a quasi-real-time framework, to unexpected events by offering multimodal transport solution to users.

Keywords: Network Management, Road Traffic Control, Theory and Models for Optimization and Control
Abstract: This paper addresses the problem of a boundary control design for traffic evolving in a large urban network. The traffic state is described on a macroscopic scale and corresponds to the vehicle density, whose dynamics are governed by a two dimensional conservation law. We aim at designing a boundary control law such that the throughput of vehicles in a congested area is maximized. Thereby, the only knowledge we use is the network's topology, capacities of its roads and speed limits. In order to achieve this goal, we treat a 2D equation as a set of 1D equations by introducing curvilinear coordinates satisfying special properties. The theoretical results are verified on a numerical example, where an initially fully congested area is driven to a state with maximum possible throughput.

Keywords: Network Management, Travel Information, Travel Guidance, and Travel Demand Management, Network Modeling
Abstract: We propose a path guidance system with a user-based charge and subsidy (UBCS) scheme for single O-D network mobility management. Users who are willing to join the scheme (subscribers) can submit travel requests along with their VOTs to the system before traveling. Those who are not willing to join (outsiders) only need to submit travel requests to the system. Our system will give all users path guidance from their origins to their destinations, and collect a path payment from the UBCS subscribers. Subscribers will be charged or subsided in a way that renders the UBCS strategy-proof, revenue-neutral, and Pareto-improving. A numerical example shows that the UBCS scheme is equitable and progressive.

Keywords: Network Management, Road Traffic Control, Data Mining and Data Analysis
Abstract: Short-term traffic forecasting is anticipated as an always evolving research topic, boosted by the tremendous recent advances of Machine Learning and Deep Learning, as well as computational power of modern PCs. In this paper, the Dilated Recurrent Neural Networks are introduced in traffic forecasting. Their architecture promotes the deployment of long-term relations and prevents common issues of RNNs, such as exploding and vanishing gradients. The Dilated LSTM Network is exploited to perform traffic conditions forecasting using network-wide data. The data consist of GPS trajectories of ride-hailing company DiDi’s vehicles from November of 2016. After preprocessing the data and organizing them into section’s travel speed of five-minute time resolution timeseries for each one of the 498 road sections of the road network of Xi’an, China, we fed them to the Dilated LSTM Network. The model consists of four hidden layers, each of them implementing an LSTM Network with one, two and four-step dilation correspondingly. The model achieves 85% accuracy, which is improved over a classic LSTM structure, trained on the same data.

Keywords: Network Management, Travel Information, Travel Guidance, and Travel Demand Management, Road Traffic Control
Abstract: This paper develops a data-driven toolkit for traffic forecasting using high-resolution (a.k.a. event-based) traffic data. This is the raw data obtained from fixed sensors in urban roads. Time series of such raw data exhibit heavy fluctuations from one time step to the next (typically on the order of 0.1-1 second). Short-term forecasts (10-30 seconds into the future) of traffic conditions are critical for traffic operations applications (e.g., adaptive signal control). But traffic forecasting tools in the literature deal predominantly with 3-5 minute aggregated data, where the typical signal cycle is on the order of 2 minutes. This renders such forecasts useless at the operations level. To this end, we model the traffic forecasting problem as a matrix completion problem, where the forecasting inputs are mapped to a higher dimensional space using kernels. The formulation allows us to capture both nonlinear dependencies between forecasting inputs and outputs but also allows us to capture dependencies among the inputs. These dependencies correspond to correlations between different locations in the network. We further employ adaptive boosting to enhance the training accuracy and capture historical patterns in the data. The performance of the proposed methods is verified using high-resolution data obtained from a real-world traffic network in Abu Dhabi, UAE. Our experimental results show that the proposed method outperforms other state-of-the-art algorithms.

Keywords: Electronic Payment
Abstract: The tolling environment in Greece is based on open zone-based tolling for inter-urban motorways, with zones ranging from 30 to 50 km, and toll fares proportional to the length of the charging zone. This is not a fully proportional distance-based scheme for users who do not travel the full length of the zone but it was not possible to construct a fully closed system with ramp toll plazas at all entries and exits, due to space and cost restrictions. The implementation of a Free Flow Electronic Toll Collection (ETC) System was not possible, mainly due to legal framework issues of enforcement and toll collectability. In order to improve the toll proportionality, Olympia Odos is combining existing conventional toll plazas with ETC gantries installed at selected entry and exit ramps resulting to a fully closed ETC or “Hybrid” Toll System, the first motorway-wide deployment of such a system in Greece and in the world. This pioneering approach to create a closed distance based ETC system, while leaving the conventional cash-based collection unaffected, is expected to be implemented also in other Greek Motorways, while it will pave the way for the future deployment of Multi Lane Free Flow ETC in Greece.

Keywords: Emission and Noise Mitigation, Network Management, Electric Vehicles
Abstract: New electrified powertrains are increasingly entering the vehicular fleet and therefore their energy response to traffic management measures that have been designed for conventional vehicles is under consideration. The present paper examines the effect of traffic signals’ adjustment on the energy consumption and CO2 emissions of various types of modern powertrains, such as hybrids and electric vehicles, estimating the effect of signal settings on vehicular energy consumption. This examination follows a modeling approach, where vehicular speed profiles for various signal setting scenarios were evaluated in energy terms, with the use of detailed instantaneous powertrain models of hybrid and electric vehicles. The evaluation reveals the formed trends on the energy performance of modern vehicles when an adjustment of traffic signal settings is applied to traffic. The recognition of these trends is essential as traffic streams will be increasingly penetrated by new electrified powertrains.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Public Transportation Management, Human Factors in Intelligent Transportation Systems
Abstract: Shared Automated Vehicles (SAVs) are a new road-based means of transport, usually small in size and capacity, with a relatively low operating speed and no (regular) possibility for the user to engage in any of the driving tasks. Past research focused on the implication of fully Automated Vehicles (AVs) in the transport sector, especially automated cars, analysing travel behaviour, network design, costs and infrastructure development. Such an extensive research on SAVs cannot be found, and most results are based on predictions for AVs acceptance instead, next to simulation studies, assumption-based models or stated choice experiments. In this paper we conduct a meta-analysis of existing literature, analysing the underlying factors that determine the adoption of SAVs. We identify the factors that have a positive effect, the ones that have a negative effect and the ones for which the effect is still unknown. Subsequently, we propose a conceptual scheme to illustrate the links between the public transport network components and the implementation of SAVs, defining a set of research questions that can help integrate SAVs in the public transport system.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Theory and Models for Optimization and Control, Simulation and Modeling
Abstract: The design of autonomous vehicles (AVs) and the design of AV-enabled mobility systems are closely coupled. Indeed, knowledge about the intended service of AVs would impact their design and deployment process, whilst insights about their technological development could signicantly affect transportation management decisions. This calls for tools to study such a coupling and co-design AVs and AV-enabled mobility systems in terms of different objectives. In this paper, we instantiate a framework to address such co-design problems. In particular, we leverage the recently developed theory of co-design to frame and solve the problem of designing and deploying an intermodal Autonomous Mobility-on-Demand system, whereby AVs service travel demands jointly with public transit, in terms of eet sizing, vehicle autonomy, and public transit service frequency. Our framework is modular and compositional, allowing one to describe the design problem as the interconnection of its individual components and to tackle it from a system-level perspective. To showcase our methodology, we present a real-world case study for Washington D.C., USA. Our work suggests that it is possible to create user-friendly optimization tools to systematically assess costs and benets of interventions, and that such analytical techniques might gain a momentous role in policy-making in the future.

Keywords: Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Simulation and Modeling, Theory and Models for Optimization and Control
Abstract: The majority of research analyzing autonomous vehicle impacts focuses on an automated, connected, electrified, and shared future. Yet, conventional private vehicle ownership and its convenience can shape a future in which autonomous vehicles (AVs) are privately-owned. Such a future would bring in a wide range of impacts on traffic flow, overall transportation system performance, energy consumption characteristics, and travel demand. In this study, we present an optimization model framework coupled with a transportation simulation tool and assess energy and mobility implications under a variety of partial- and full-automation scenarios with different costs and states of the technology assumptions. We demonstrate that private AVs significantly influence the regional traffic congestion of a test area (Bloomington, Illinois) with a 74% increase in vehicle hours traveled (VHT) and 66% and 87% increase in fuel consumption under low and high technology adoption scenarios, respectively. However, we find that introducing 0.10 taxation per mile unoccupied travel could mitigate the influence by reducing VHT and fuel consumption 13% and 15-20%, respectively.

Keywords: Air Traffic Management, Aerial, Marine and Surface Intelligent Vehicles, Cooperative Techniques and Systems
Abstract: With increasing urban population, there is global interest in Urban Air Mobility (UAM), where hundreds of autonomous Unmanned Aircraft Systems (UAS) execute missions in the airspace above cities. Unlike traditional human-in-the-loop air traffic management, UAM requires decentralized autonomous approaches that scale for an order of magnitude higher aircraft densities and are applicable to urban settings. We present Learning-to-Fly (L2F), a decentralized on-demand airborne collision avoidance framework for multiple UAS that allows them to independently plan and safely execute missions with spatial, temporal and reactive objectives expressed using Signal Temporal Logic. We formulate the problem of predictively avoiding collisions between two UAS without violating mission objectives as a Mixed Integer Linear Program (MILP). This however is intractable to solve online. Instead, we develop L2F, a two-stage collision avoidance method that consists of: 1) a learning-based decision-making scheme and 2) a distributed, linear programming-based UAS control algorithm. Through extensive simulations, we show the real-time applicability of our method which is ~6000x faster than the MILP approach and can resolve 100% of collisions when there is ample room to maneuver, and shows graceful degradation in performance otherwise. We also compare L2F to two other methods and demonstrate an implementation on quad-rotor robots.

Keywords: Air Traffic Management, Data Mining and Data Analysis
Abstract: As an essential part of a flight life cycle, the surface taxiing process is closely related to the operational efficiency of the airport. Routing and scheduling can be optimized with an accurate prediction of aircraft taxi time in advance, thus improving the ability of refined management of airport surface. However, the past methods merely provide a taxi time predicted by their models, which are of limited help to airport controllers in decision-making. In order to alleviate this problem, this paper proposes to use a deep metric learning (DML) model to learn the similarity between historical operation scenarios based on basic flight properties, surface traffic situation, and airport weather information. For a given reference flight, the taxi time can be reasonably predicted by finding its similar historical scenarios. In this way, the relevant controllers can make flexible decisions at the tactical level. Experimental verification on the historical data of Shanghai Pudong International Airport shows that the deep model can effectively capture the similarity of taxi time between different scenarios. Besides, compared with the classical machine learning prediction models, the proposed model can predict the taxi time more accurately. With similar historical scenarios as the basis for decision support, the implementation and interpretability of Airport Collaborative Decision-Making (A-CDM) system will be enhanced.

Keywords: Sensing, Vision, and Perception, Roadside and On-board Safety Monitoring, Advanced Vehicle Safety Systems
Abstract: In this paper, we propose a novel deep learning architecture for the end-to-end classification of unsafe maneuvers from dashcam data; the proposed model is based on an innovative two-stream architecture capable of processing both video and GPS/IMU signals as input streams. A wide experimentation on a well known naturalistic driving dataset (SHRP2 NDS) shows that the two sources of information complement each other in the classification task and proves the effectiveness of the proposed approach. As a by-product of this research, we propose and make available a novel classification of safety-critical events based on the unsafe maneuver leading to them, which is representative of the real distribution of car crashes and near crashes.

Keywords: Sensing, Vision, and Perception, Multi-modal ITS, Other Theories, Applications, and Technologies
Abstract: In this paper, we propose deepRGBXYZ – a feature descriptor to represent pixels for robust dense pixel matching. To this end, we concatenate RGB image (appearance) information with the depth (geometric) information represented as XYZ in order to build a robust descriptor which is more invariant to photometric and geometric changes. Both information (RGB and depth) are embedded as an early fusion into one neural network which is based on stacked dilated convolutions for enlarging the receptive field. We alleviate the limitations of image-only descriptors especially within ill-conditioned light regions or textureless objects. Additionally, we overcome the difficulty of using depth-only information which show less descriptive details compared to image-only. We demonstrate the superior accuracy of our deepRGBXYZ descriptor against the state-of-the-art image-only descriptors and we verify our design decision. In addition, we investigate the superior robustness of our deepRGBXYZ descriptor by bringing it into the application of optical flow and scene flow estimation on the established data sets KITTI and FlyingThings3D.

Keywords: Sensing, Vision, and Perception, Driver Assistance Systems
Abstract: Automotive cameras, particularly surround-view cameras, tend to get soiled by mud, water, snow, etc. For higher levels of autonomous driving, it is necessary to have a soiling detection algorithm which will trigger an automatic cleaning system. Localized detection of soiling in an image is necessary to control the cleaning system. It is also necessary to enable partial functionality in unsoiled areas while reducing confidence in soiled areas. Although this can be solved using a semantic segmentation task, we explore a more efficient solution targeting deployment in low power embedded system. We propose a novel method to regress the area of each soiling type within a tile directly. We refer to this as coverage. The proposed approach is better than learning the dominant class in a tile as multiple soiling types occur within a tile commonly. It also has the advantage of dealing with coarse polygon annotation, which will cause the segmentation task. The proposed soiling coverage decoder is an order of magnitude faster than an equivalent segmentation decoder. We also integrated it into an object detection and semantic segmentation multi-task model using an asynchronous back-propagation algorithm. A portion of the dataset used will be released publicly as part of our WoodScape dataset cite{yogamani2019woodscape} to encourage further research.

Keywords: Sensing, Vision, and Perception
Abstract: Predicting the condition of the road is an important task for autonomous vehicles to make driving decisions. Vehicles are expected to slow down or stop for potential road risks such as road cracks, bumps and potholes. Vision systems are widely used to provide such information given the rich colours and textures carried by images. This paper presents a weakly-supervised deep learning method to classify road images into two category sets. The first category identifies the existence of bumps or ramps in the image. The second category determines the road roughness given an input image. These two outputs are combined into a single convolutional neural network (CNN) to classify the camera image simultaneously. As a supervised learning method, deep learning algorithms normally require a large amount of training data with manually annotated labels. The annotation process is, however, very time-consuming and labour-intensive. This paper presents a method to avoid this costly process using a pipeline to automatically generate ground-truth labels by incorporating IMU and wheel encoder data. This automated pipeline does not require human effort to label images and will not be impeded by adverse environmental or illumination conditions. The experimental results presented show that after training the model using the automatically generated labels, the two-output CNN is capable to achieve good accuracy for classifying road conditions.

Keywords: Sensing, Vision, and Perception, Advanced Vehicle Safety Systems, Sensing and Intervening, Detectors and Actuators
Abstract: Object detection algorithms for Lidar data have seen numerous publications in recent years, reporting good results on dataset benchmarks oriented towards automotive requirements. Nevertheless, many of these are not deployable to embedded vehicle systems, as they require immense computational power to be executed close to real time. In this work, we propose a way to facilitate real-time Lidar object classification on CPU. We show how our approach uses segmented object instances to extract important features, enabling a computationally efficient batch-wise classification. For this, we introduce a data representation which translates three-dimensional information into small image patches, using decomposed normal vector images. We couple this with dedicated object statistics to handle edge cases. We apply our method on the tasks of object detection and semantic segmentation, as well as the relatively new challenge of panoptic segmentation. Through evaluation, we show, that our algorithm is capable of producing good results on public data, while running in real time on CPU without using specific optimisation.

Keywords: Simulation and Modeling, Sensing, Vision, and Perception, Data Mining and Data Analysis
Abstract: In this work, we investigate the potential of latent representations generated by Variational Autoencoders (VAE) to analyze and distinguish between real and synthetic data. Although the details of the domain adaptation task are not the focus of this work, we use the example of simulated lidar data adapted by a generative model to match real lidar data. To assess the resulting adapted data, we evaluate the potential of latent representations learned by a VAE. During training, the VAE aims to reduce the input data to a fixed-dimensional feature vector, while also enforcing stochastic independence between the latent variables. These properties can be used to define pseudometrics to make statements about generative models that perform domain adaptation tasks. The variational autoencoder is trained on real target data only and is subsequently used to generate distributions of feature vectors for data coming from different data sources such as simulations or the output of Generative Adversarial Networks.

Keywords: Simulation and Modeling, Theory and Models for Optimization and Control, Cooperative Techniques and Systems
Abstract: OmniCAV is laying the foundations for the development of a comprehensive, robust and secure simulator, aimed at providing a certification tool for Connected Autonomous Vehicles (CAVs) that can be used by regulatory and accreditation bodies, insurers and manufacturers to accelerate the safe development of CAVs. To achieve this, OmniCAV is using highly detailed road maps, together with a powerful combination of traffic management, accident and CCTV data, to create a high-fidelity traffic and driving simulation environment to interact with the AV under test. Scenarios for testing are developed and randomised in a holistic way to avoid CAVs training to specific conditions. Critically, the simulator offers coverage of a representative element of the U.K. road network, through encompassing rural roads, peri-urban and urban roads to enable autonomy for all. The validity of the synthetic test environment compared to the real-world is of particular importance, and OmniCAV will be tested and refined through an iterative approach involving real-world comparisons and working in conjunction with a CAV testbed.

Keywords: Simulation and Modeling, Driver Assistance Systems, Theory and Models for Optimization and Control
Abstract: Motion planning is a major component of any automated driving system. The safety assessment of such components requires a formal characterization of the perception and control mechanisms. This requires dedicated tools and models for the environment, sensors and vehicles that are highly representative of the real world. Simulation is a method to virtually investigate the behavior of systems under study. It has a key role to play in demonstrating the safety of autonomous vehicles. In this context, we consider a control module as a black-box and try to determine a reference which represents the ’right decision’, if it exists. An optimization-based reference model is created for the control function. This model allows each scene in the environment to be mapped to the desired decision regardless of the black-box. The black-box and the reference model are run on several critical scenarios. In output, an assessment of decision making is performed along with systematic criticality characterization of targeted scenarios.

Keywords: Simulation and Modeling, Data Mining and Data Analysis, Electric Vehicles
Abstract: Synthetic Driving Cycles have been used in numerous studies to describe a certain driving profile of relevance. An important purpose of synthetic cycles is to limit the necessary time on a test-rig or to reduce the computational effort within simulations, which is achieved by compressing a larger amount of gathered operating data from a certain vehicle or a vehicle fleet to a necessary minimum. Interestingly, despite the intensive use of the synthetic driving cycles, there is only limited literature on the validation of using synthetic driving cycles. Therefore, the scope of this work is to further investigate under which conditions synthetic driving cycles can be used to replace the entirety of the relevant operating data in the evaluation of a vehicle’s consumption. We apply a longitudinal vehicle simulation model to calculate the fuel and electric consumption of vehicles with different powertrain concepts on many generated synthetic driving cycles for different compression rates. We then compare that to the consumption if considering the original driving data. A legislative driving cycle (WLTC) as well as naturalistic driving data sets are used for the evaluation. The results show, that synthetic driving cycles allow for a compact representation of the original data sets but possible compression rates depend on the specific driving data. The presented two-step process can be extended to a generalized validation process for the use of synthetic driving cycles.

Keywords: Simulation and Modeling, Theory and Models for Optimization and Control, Automated Vehicle Operation, Motion Planning, Navigation
Abstract: This paper provides a nonlinear observer for vehicle motion tracking system in lane change manoeuvre. Firstly, the vehicle body axis system is designed and the nonlinear vehicle motion dynamics is then studied. By using differential mean value theorem and introducing vehicle’s uncertainties, a nonlinear observer is designed and a estimation error system is developed. Applying the Lyapunov theory and Linear MatrixInequality (LMI), the designed observer gains are obtained.Stability, effectiveness and potency of the theoretical results are confirmed in lane change manoeuvre by the simulation results

Keywords: Other Theories, Applications, and Technologies, Travel Behavior Under ITS, Travel Information, Travel Guidance, and Travel Demand Management
Abstract: In this paper, we address the much-anticipated deployment of connected and automated vehicles (CAVs) in society by modeling and analyzing the social-mobility dilemma in a game-theoretic approach. We formulate this dilemma as a normal-form game of players making a binary decision for their mobility and construct an intuitive payoff function inspired by the socially beneficial outcomes of a mobility system consisting of CAVs. We show that the game is equivalent to the Prisoner's dilemma, which implies that the rational collective decision is the opposite of the socially optimum. We present two different solutions to tackle this phenomenon: one with a preference structure and the other with institutional arrangements. We conclude by showcasing our last result with a numerical study.

Keywords: Other Theories, Applications, and Technologies
Abstract: Automatic license plate detection has always been a popular topic in intelligent transport systems. In recent years, many approached the problem using artificial intelligence and machine learning techniques. Sufficient amount of good quality data is critical for machine learning, however, most of the existing open source license plate datasets are either restrictive in nature or do not have enough number of images to allow training a robust license plate detection network. In this paper, the THI License Plate Dataset (TLPD) is presented. It has more than 17,000 vehicle images and 18,000 labelled license plates in it. It is one of the largest publicly available European Union license plate datasets. The images in the dataset capture license plates at different angles and distances in relation to the camera. They are also taken under different illumination and weather conditions, making the dataset suitable for training robust license plate detectors that could work in various scenarios. A fast labelling tool for single class labelling is also presented in this paper. This tool was used to label the license plates in TLPD. This paper also introduces two license plate detection networks trained with data from TLPD. The networks work robustly under various scenarios, achieving precision and recall value of 93% under day and night conditions and 87% under heavy snowing condition.

Keywords: Other Theories, Applications, and Technologies, Advanced Vehicle Safety Systems, Roadside and On-board Safety Monitoring
Abstract: Cut-in situations occurs when a vehicle intentionally changes lane and ends up in front of another vehicle or in-between two vehicles. In such situations, having a method to indicate the collision risk prior to making the cut-in maneuver could potentially reduce the number of sideswipe and rear end collisions caused by the cut-in maneuvers. This paper propose a new risk indicator, namely cut-in risk indicator (CRI), as a way to indicate and potentially foresee collision risks in cut-in situations. As an example use case, we applied CRI on data from a driving simulation experiment involving a manually driven vehicle and an automated platoon in a highway merging situation. We then compared the results with time-to-collision (TTC), and suggest that CRI could correctly indicate collision risks in a more effective way. CRI can be computed on all vehicles involved in the cut-in situations, not only for the vehicle that is cutting in. Making it possible for other vehicles to estimate the collision risk, for example if a cut-in from another vehicle occurs, the surrounding vehicles could be warned and have the possibility to react in order to potentially avoid or mitigate accidents.

Keywords: Other Theories, Applications, and Technologies, Sensing and Intervening, Detectors and Actuators
Abstract: The effectiveness of machine learning (ML) approaches for machine fault diagnosis (MFD) has been proved in previous studies. The majority of the previous studies used simulation or laboratory datasets. However, the robustness of the ML models for (MFD) in real-world applications has rarely been discussed. In this work, we conducted an empirical study on the robustness of ML approaches in case of wheel flat (WF) detection for railway freight wagons. WF is a common failure on wheel tread surface, causing large impulsive impacts on vehicles and infrastructure. We made great efforts to collect and clean up the relevant field data that was measured on different freight wagons running with different faulty conditions on different lines at different speed ranges. The collected datasets can represent the complexity of real-world railway operational conditions. We build several baseline models, incorporating deep convolutional neural network (CNN) and different signal processing techniques for WF detection, in order to study their robustness against diverse conditional variations in practice.

Keywords: Other Theories, Applications, and Technologies, Aerial, Marine and Surface Intelligent Vehicles, Electric Vehicles
Abstract: This paper proposes a methodology of climbing and descending stairs for four-wheeled vehicles. Conventional vehicles commonly use a crawler mechanism for moving on stairs. In contrast, our vehicle is based on a wheel mechanism. The stair-climbing gait (SCG) for our vehicle is proposed in another paper; however, movement with this gait requires a great deal of time. In this paper, two motions are switched between for our vehicle to climb and descend stairs. The other motion besides SCG is the double wheel motion (DWM), in which the vehicle moves on stairs using wheels with an appropriate center of gravity position and appropriate wheel velocities. The proposed methodology is realized and evaluated through an experiment. (Please also see the attached video.)

Keywords: Ride Matching and Reservation, Simulation and Modeling, Data Mining and Data Analysis
Abstract: On-demand mobility services form a major section of the transport system, especially in metropolitan cities. Contemporary on-demand systems work in a partially coordinated manner where the drivers follow their own passenger finding strategies. This leads to sub-optimal performance at the company level due to a Spatio-temporal supply-demand mismatch. A coordinated fleet of taxis that also employs the knowledge of predicted future demand is being seen as a promising upgrade to contemporary systems. However, for this system to work, the drivers must see the clear benefits of switching from the contemporary system to the coordinated system. In other terms, the drivers must feel that the system is fair. In this paper, we have built a framework to simulate an on-demand taxi system working in a coordinated manner. Our objective is to ensure company profit maximization while improving service fairness in terms of driver profit disparity reduction as well as increasing the scheduling guarantee for drivers who are following the recommendation of the system. We show that the proposed model can serve 49.8% more demand with 71% higher mean driver profit, 83% lower profit disparity and a 39.8% higher scheduling guarantee.

Keywords: Ride Matching and Reservation, Simulation and Modeling
Abstract: We study the curbside stopping activities of on-demand mobility services from a macroscopic perspective. We develop a queueing network model of the curbside stopping system and integrate this model into existing macroscopic analyzing frameworks of on-demand mobility services. Theoretical analysis and numerical experiments both show that service operators can provide services at a level beyond the social optimum and introduce negative externalities to the system. Moreover, improvements in services' operational efficiencies are not sufficient to resolve this problem. Therefore, it is necessary to introduce system-level solutions such as introduction of additional curbside stopping spaces.

Keywords: Ride Matching and Reservation, Theory and Models for Optimization and Control, Simulation and Modeling
Abstract: Car-sharing is emerging as an effective and flexible alternative to car ownership in many large cities in the world. However, car-sharing is still dominated by single-occupancy vehicles, and its convenience may attract still more passengers from public transport, thus aggravating congestion and pollution. Ride sharing can mitigate this problem by pooling together passengers doing similar trips, thus decreasing the number of cars on the road. In this work we propose a strategy for offering ride-sharing in one-way car-sharing systems. To this end, we design a graph-based approach to model the trip-matching problem in a car-sharing system using passenger pooling. Then, we formulate both online and offline trip-matching algorithms, and we investigate their on the performance of the system using a large data set from the city of New York. We consider both the point of view of the operator and the users, in terms of net revenues and utility, respectively. We show that with the right fare structure a ride-sharing option can lead to a Pareto-optimal solution that increases both net revenues and passenger utility while decreasing vehicles-km travelled.

Keywords: Ride Matching and Reservation, Network Modeling, Simulation and Modeling
Abstract: Ride pooling services have the possibility to enhance the traffic efficiency in our cities. The improvements on traffic conditions among other factors depend on the likelihood to find shareable trips in an operational area, a quantity called shareability. There are already available mathematical models, which show that shareability depends on service quality and city parameters, such as average velocity, which is considered to be constant. However, with the introduction of ride pooling services the average velocity is expected to increase as a result of fewer vehicles on the road due to shared trips. Current models are not able to estimate the modified shareability values for a dynamic velocity, which depends on the percentage of shared trips in an area. Therefore, this paper explores analytically the traffic impacts of ride pooling services and the influence that changed traffic conditions due to ride pooling have on the percentage of shared trips. The results could be beneficial for operators and cities for a quick estimation of the traffic impacts of ride pooling and the shareability that can be reached in an area due to these effects.

Keywords: Simulation and Modeling, Ride Matching and Reservation, Network Modeling
Abstract: This paper addresses the problem of dynamically matching automated vehicles (AVs) to open traveler requests in a large-scale automated-mobility-on-demand (AMOD) simulation framework. While optimization-based matching strategies based on the linear assignment problem formulation significantly outperform simple heuristic strategies (e.g. nearest neighbor), the scalability of the assignment problem limits its applicability to large problem instances. This study proposes a fleet dispatching strategy to dynamically assign AVs to travelers that involves the assignment problem formulation but restricts the decision space to reduce computational time. First, we significantly trim the decision space via only considering the k-closest open requests around each idle vehicle or k-closest idle vehicles around each open request. Second, we only calculate point-to-point shortest paths for vehicles and travelers that are close in spatial proximity. For vehicles and travelers that are not close in proximity, we use zone-to-zone travel time estimates. This study embeds the proposed AV fleet dispatching strategy within Polaris--an agent-based transportation simulation modeling framework. Within Polaris, the restricted fleet dispatching strategy proposed in this significantly outperforms (i) existing large-scale strategies in terms of fleet performance and (ii) the unrestricted assignment problem strategy in terms of computational performance.

Keywords: Travel Behavior Under ITS, Transportation Security, Data Mining and Data Analysis
Abstract: Driving confidence psychology can guide drivers in calm driving operation when dealing with traffic issues, which is of substantial significance for reducing the accident rate and improving the road traffic efficiency. This study mainly analyzes the differences in driving confidence psychology in the face of an emergency braking event of a front vehicle with warning as opposed to the same situation without warning information. First, an emergency braking event of a front vehicle in a connected vehicle environment was designed based on driving simulation technology, which can provide warning information from the emergency-braking vehicle by using an onboard human-machine interface (HMI). Second, the features of lateral lane position changing and the average angle of the gas pedal were used to analyze the differences in driving confidence with versus without warning information. Finally, the entropy weight method was used to obtain the driving confidence degree of each driver in both scenarios. The results demonstrate that the driving confidence level is higher when warning information is provided, and the average driving confidence degree is 2.11% higher than the average driving confidence degree without warning information.

Keywords: Travel Behavior Under ITS, Data Mining and Data Analysis
Abstract: Occupancy of vacant taxis brings waste of resources and heavy traffic pressure to urban traffic network. To enable better efficiency of taxis and less traffic congestion, a logit-based model is developed to describe vacant taxi drivers’ route choice behavior and decision-making mechanism when searching for next customer. The proposed model is based on a multinomial logit model (MNL), considering two novel influencing indicators including the path unreliability (PU) and expected rate of return (EROR). In order to simplify the model construction and reduce computational cost, we divide the research area into identical squares with a 0.5-km resolution. Then the customer searching movements are extracted from the high-resolution GPS data of more than 8000 taxis in Shanghai to validate the logit model. The model results show that the customer searching behavior of vacant taxi drivers is significantly influenced by PU and EROR. Moreover, the effect of these two indicators on customer searching behavior varies with the time of day.

Keywords: Travel Behavior Under ITS, Incident Management
Abstract: Intersections are one of the main sources of congestion and hence, it is important to understand traffic behavior at intersections. Particularly, in developing countries with high vehicle density, mixed traffic type, and lane-less driving behavior, it is difficult to distinguish between congested and normal traffic behavior. In this work, we propose a way to understand the traffic state of smaller spatial regions at intersections using traffic graphs. The way these traffic graphs evolve over time reveals different traffic states - a) a congestion is forming (clumping), the congestion is dispersing (unclumping), or c) the traffic is flowing normally (neutral). We train a spatio-temporal deep network to identify these changes. Also, we introduce a large dataset called EyeonTraffic (EoT) containing 3 hours of aerial videos collected at 3 busy intersections in Ahmedabad, India. Our experiments on the EoT dataset show that the traffic graphs can help in correctly identifying congestion-prone behavior in different spatial regions of an intersection.

Keywords: Travel Behavior Under ITS, Multi-modal ITS, Travel Information, Travel Guidance, and Travel Demand Management
Abstract: This paper studies congestion-aware route-planning policies for Autonomous Mobility-on-Demand (AMoD) systems, whereby a fleet of autonomous vehicles provides on-demand mobility under mixed traffic conditions. Specifically, we first devise a network flow model to optimize the AMoD routing and rebalancing strategies in a congestion-aware fashion by accounting for the endogenous impact of AMoD flows on travel time. Second, we capture reactive exogenous traffic consisting of private vehicles selfishly adapting to the AMoD flows in a user-centric fashion by leveraging an iterative approach. Finally, we showcase the effectiveness of our framework with a case-study considering the transportation sub-network in New York City. Our results suggest that for high levels of demand, pure AMoD travel can be detrimental due to the additional traffic stemming from its rebalancing flows, whilst the combination of AMoD with walking or micromobility options can significantly improve the overall system performance.

Keywords: Theory and Models for Optimization and Control, Travel Behavior Under ITS
Abstract: This paper proposes a new method for the determination of optimal control strategy with considering multiple time constraints. For the train on a level track, the optimal train control model is firstly formulated. Then, with the given control sequences, Kuhn-Tucker condition is used to deduce the necessary conditions for a strategy of optimal type and an analytical solution with the minimum energy consumption is proposed to calculate the optimal control strategy which contains the specific switching points for the given control sequence. Finally, two realistic examples will be applied to illustrate the numerical calculation procedures and prove the effectiveness of the proposed approach.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems, Driver Assistance Systems
Abstract: We propose an approach for the fast computation of reachable sets of road vehicles while considering dynamic obstacles. The obtained reachable sets contain all possible behaviors of vehicles and can be used for motion planning, verification, and criticality assessment. The proposed approach precomputes computationally expensive parts of the reachability analysis. Further, we partition the reachable set into cells and construct a directed graph storing which cells are reachable from which cells at preceding time steps. Using this approach, considering obstacles reduces to deleting nodes from the directed graph. Although this simple idea ensures an efficient computation, the discretization can introduce considerable over-approximations. Thus, the main novelty of this paper is to reduce the over-approximations by intersecting reachable sets propagated from multiple points in time. We demonstrate our approach on a large range of scenarios for automated vehicles showing a faster computation time compared to previous approaches while providing the same level of accuracy.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Multi-autonomous Vehicle Studies, Models, Techniques and Simulations, Driver Assistance Systems
Abstract: Motion planning for an autonomous vehicle is most challenging for scenarios such as large, multi-lane, and unsignalized intersections in the presence of dense traffic. In such situations, the motion planner has to deal with multiple crossing-points to reach an objective in a safe, comfortable, and efficient way. In addition, motion planning challenges include real-time computation and scalability to complex scenes with many objects and different road geometries. In this work, we propose a motion planning system addressing these challenges. We enable real-time applicability of a Monte Carlo Tree Search algorithm with a deep-learning heuristic. We learn a fast evaluation function from accurate, but non real-time models. While using Deep Reinforcement Learning techniques we maintain a clear separation between making predictions and making decisions. We reduce the complexity of the search model and benchmark the proposed agent against multiple methods: rules-based, MCTS, A* search, deep learning, and Model Predictive Control. We show that our agent outperforms these other agents in a variety of challenging scenarios, where we benchmark safety, comfort and efficiency metrics.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Cooperative Techniques and Systems, Road Traffic Control
Abstract: We address the problem of optimally controlling Connected and Automated Vehicles (CAVs) arriving from two multi-lane roads and merging at multiple points where the objective is to jointly minimize the travel time and energy consumption of each CAV subject to speed-dependent safety constraints, as well as speed and acceleration constraints. This problem was solved in prior work for two single-lane roads. A direct extension to multi-lane roads is limited by the computational complexity required to obtain an explicit optimal control solution. Instead, we propose a general framework that converts a multi-lane merging problem into a decentralized optimal control problem for each CAV in a less-conservative way. To accomplish this, we employ a joint optimal control and barrier function method to efficiently get an optimal control for each CAV with guaranteed satisfaction of all constraints. Simulation examples are included to compare the performance of the proposed framework to a baseline provided by human-driven vehicles with results showing significant improvements in both time and energy metrics.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation
Abstract: Seamlessly operating an autonomous vehicles in a crowded pedestrian environment is a very challenging task. This is because human movement and interactions are very hard to predict in such environments. Recent work has demonstrated that reinforcement learning-based methods have the ability to learn to drive in crowds. However, these methods can have very poor performance due to making inaccurate predictions of the pedestrians’ future state as human motion has a large variance. To overcome this problem, we propose a new method, SARL-SGAN-KCE, that combines deep socially aware attentive value network with a human multimodal trajectory prediction model to help identify the optimal driving policy. We also introduce a novel technique to extend the discrete action space with little extra computational requirements. The kinematic constraints of the vehicle are also considered to ensure smooth and safe trajectories. We evaluate our method against the state of art methods for crowds navigation and provide an ablation study to show that our method is safer and closer to human behaviour.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems, Driver Assistance Systems
Abstract: Avoiding critical situations is a prerequisite for Advanced Driver Assistant Systems and Autonomous Driving to decrease the number of total hazards and fatal collisions. As a guide for safe motion behavior and for avoiding critical situations in complex scenarios with several interacting traffic participants, an appropriate risk measurement is necessary. It should incorporate system-inherent uncertainties like present in environment recognition, behavior predictions and physical model assumptions. In this paper, we introduce a time-course-aware incremental risk model for motion planning which predicts state distributions along forecasted trajectories and regards their magnitude evolution by the Survival Theory and their shape adaptation by removing collided distribution parts while preserving statistical moments. Our approach is able to reproduce motion risk probability costs as found by particle-based Monte-Carlo (MC) simulations in a range of scenarios, at much lower computational costs.

Keywords: Data Mining and Data Analysis, Travel Behavior Under ITS
Abstract: Leveraging the crawled geotagged and timestamped tweets of Twitter users, this study develops a methodological framework to predict massively unreported travel mode choices of Twitter users who have left geotagged and timestamped tweets. The prediction framework is based on the similarity between a user who has not reported her mode choice and the users with known travel modes. To appropriately represent a Twitter user’s data, we employ a discretized spatial-temporal probabilistic distribution to characterize the user. A novel convolution-based similarity measure is then proposed to effectively capture the interdependencies of both spatially and temporally adjacent data points. A graph inference model is further established to explore the predictability of people’s travel mode choice. To validate the prediction framework, we use the Proposition 1 incident in Austin, TX in 2016 as a case study and leverage relevant data crawled from Twitter. The prediction results validate the effectiveness of both the convolution-based similarity measure and the prediction framework. This work demonstrates the feasibility of using social media data to predict people’s mobility choices.

Keywords: Data Mining and Data Analysis
Abstract: The future of transportation is tightly connected to Autonomous Driving (AD). While a lot of progress has been made in recent years, there are still obstacles to overcome. One of the most critical issues is the safety verification of AD. A scenario-based verification approach that shifts tests from the fields to a virtual environment seems like a sophisticated approach to tackle the safety verification as tests need to be revised whenever changes are made to the AD. However, collecting and labelling data that can be used to construct scenarios is expensive and time-consuming to compute. In this work, we propose a unified framework for trajectory generation and validation in a consistent and principled way. We first explore methods to generate artificial trajectories that resemble the previously captured ones. More specifically, we consider two architectures based on Generative Adversarial Networks (GANs): recurrent GANs and a recurrent Autoencoder in combination with GANs. Moreover, we investigate the use of different metrics to evaluate the quality of generated trajectories which is a nontrivial task.

Keywords: Data Mining and Data Analysis, Theory and Models for Optimization and Control, Simulation and Modeling
Abstract: Cruising for parking is not only stressful task for most drivers but also increases congestion and emissions. Therefore smart parking guidance systems are gaining increasing interest from researchers and city councils. These systems mostly rely on expensive and not well scalable technology like real time parking sensors or camera systems. In this paper we propose a deep learning architecture that predicts the current number of parking cars at different locations based on digital meter payment transactions. We outperform simple baseline models as well as a state of the art probabilistic approach from the literature. Transactional data does not directly translate to parking occupancy since not all people stick to their paid duration or pay at all. We therefore discuss the reliability of our method on different datasets and spatial granularities. Although our model is not as reliable as sensor data, especially for small parking zones, we find that our methodology provides an inexpensive way of inferring on-street parking occupancy and enable meaningful smart parking services.

Keywords: Data Mining and Data Analysis
Abstract: In traditional methods, fundamental diagrams (FDs) were calibrated offline with a limited number of links. Although few recent studies have paid attention to employing cluster techniques to calibrate link FDs for network level analysis, they were mainly focused on heuristic clustering methods, such as k-means and hierarchical clustering algorithm which might lead to poor performance when there are overlaps between clusters. This paper proposed a mixture model-based clustering framework to calibrate link FDs for network level simulation. This method can be applied to discover a relatively small number of representative link FDs when simulating very large networks with time and budget constraints. In addition, the proposed method can be used to investigate the spatial distribution of links with similar FDs. The proposed method is tested with 567 links using one year’s data from the Northern California.

Keywords: Data Mining and Data Analysis, Data Management and Geographic Information Systems, Driver Assistance Systems
Abstract: The development of Advanced Driver Assistance Systems (ADAS) is proceeding rapidy, leading to fastly growing pools of recorded driving data for testing and validation. To make use of this data pool efficiently, a low barrier access for development engineers is of high importance, facilitating reuse of the data for function assessment or simulation. The main contribution of this work is the adoption of ideas from Natural Language Processing (NLP) to enable the semantic interpretation of driving data in their respective contexts, facilitating search and comparison in the data pool. The suggested process consists of a binning to convert the multivariate time series into discrete drive states and a subsequent transformation into a lower-dimensional space, a so-called embedding. The transformation is performed by the adaptation of Word2Vec, a model originally developed for the embedding of words in language processing tasks. The resulting drive state embedding can be used to compare driving sequences on a semantic level with the inclusion of their contexts, permitting applications like the extraction of similar sequences or the detection of anomalous events.

Keywords: Travel Information, Travel Guidance, and Travel Demand Management, Simulation and Modeling, Road Traffic Control
Abstract: In this paper, we jointly integrate route-reservations with a pricing mechanism to evaluate the effect of congestion pricing on the driver departure time choices. Route-reservations have shown to be a durable congestion mitigation mechanism that can achieve up to 70% reduction in travel times. Unfortunately, this improvement is achieved only when the majority of the drivers comply with the suggested routes and departure times. Therefore, a pricing mechanism is proposed that allows drivers to deviate from the suggested departure times. To identify the departure time choices of drivers we explicitly take into account their desired departure time from their origin and also the start time of the activity they are planning to perform. The proposed flexible route-reservation framework is evaluated in a microscopic simulation with results demonstrating how the introduced pricing mechanism can eliminate congestion while allowing flexibility to drivers to deviated from the suggested departure time.

Keywords: Travel Information, Travel Guidance, and Travel Demand Management, Data Mining and Data Analysis, Accurate Global Positioning
Abstract: Location tracking is an essential problem for mobility-based applications that facilitate the daily life of Smartphone users. Existing applications often use energy-hungry sensors like GPS or gyroscope to detect significant journeys. Recent research has often focused on optimizing energy consumption. As a result, approaches were proposed using sensors fusions, hybrid or eventual sensors selection. However, such research largely neglects the performance in underground tracking of automotive mobility. Possible solutions, such as those involving barometers, have well-known issues regarding performance. Oppositely, although energy-friendly, accelerometers are often overlooked based on the assumption that pattern extraction is hard due to over-noisy characteristics of the signal. In this paper, we propose a ready-to-use Framework for underground train tracking. This Framework uses an adaptive Singular Spectrum Analysis (SSA) to process the Accelerometer data. We run an empirical study using data collected from Smartphone embedded accelerometers, to track departings and arrivals of the trains in four large European cities. Results show that: 1) the Framework is able to accurately locate the trains; 2) SSA adds improvements compared to Butterworth filters and complementary filter with sensors fusion.

Keywords: Travel Information, Travel Guidance, and Travel Demand Management, Theory and Models for Optimization and Control, Network Modeling
Abstract: Optimal routing in urban transit networks, where variable congestion levels often lead to stochastic travel times, is usually studied with the least expected travel time (LET) as the performance criteria under the assumption of travel time independence on different road segments. However, a LET path might be subjected to high variability of travel time and therefore might not be desirable to transit users seeking a predictable arrival time. Further, there exists a spatial correlation in urban travel times due to the cascading effect of congestion across the road network. In this work, we propose a methodology and a tool that, given an origin-destination pair, a travel time budget, and a measure of the passenger's tolerance for uncertainty, provide the optimal online route choice in a transit network by balancing the objectives of maximizing on-time arrival probability and minimizing expected travel time. Our framework takes into account the correlation between travel time of different edges along a route and updates downstream distributions by taking advantage of upstream real-time information. We demonstrate the utility and performance of our algorithm with the help of realistic numerical experiments conducted on a fixed-route bus system that serves the residents of the Champaign-Urbana metropolitan area.

Keywords: Travel Information, Travel Guidance, and Travel Demand Management, Modeling, Simulation, and Control of Pedestrians and Cyclists
Abstract: The travel planning system is an important reference tool for passengers to choose a travel route. An efficient and safe travel plan can improve the travel efficiency and ride safety of passengers, and improve the service level of the rail transit system. At the same time, travel planning information is of great value for more accurate prediction of short-term passenger flow, and more accurate short-term passenger flow prediction information can also provide support for higher-quality travel planning. Existing related studies have hardly considered the information interaction mechanism between passenger flow predicting and travel planning. To solve this problem, this paper proposes a new regional rail transit travel planning system design idea and implementation scheme based on passenger flow prediction. This system makes short-term prediction of passenger flow based on individual travel planning information. The passenger flow prediction results are used to provide support for subsequent passenger travel planning. Finally, the Chongqing rail transit network is taken as an example to analyze the prototype case, and the feasibility of the system scheme was preliminarily verified.

Keywords: Travel Information, Travel Guidance, and Travel Demand Management, Data Mining and Data Analysis
Abstract: Understanding how people choose to travel is essential for intelligent transportation planning and related smart services. Recent advances in deep learning, coupled with the increasing market penetration of GPS devices, have paved the way for novel travel mode identification methods based on GPS data mining. While many have shown promising results, most methods have often relied heavily on the few available labeled data, leaving large amounts of unlabeled ones unused. To address this issue, we propose MultiMix, a semi-supervised multi-task learning framework for travel mode identification. Our framework trains a deep autoencoder using batches of labeled, unlabeled, and synthetic data by simultaneously optimizing three corresponding objective functions. We show that MultiMix outperforms several fully- and semi-supervised baselines, achieving a classification accuracy of 66.2% on Geolife using just 1% of labeled data, with accuracy reaching 84.8% when incorporating all available labels. We also verify the necessity of its components through an ablation study designed to provide insights into the proposed approach.

Keywords: Simulation and Modeling, Data Mining and Data Analysis, Rail Traffic Management
Abstract: In order to provide a satisfying quality of service to passengers, companies operating suburban trains seek to implement timetables that perform well despite the occurrence of random disturbances. However, due to some specificities of Mass Transit, the standard approaches for robust train timetabling do not apply in that context. In this paper, we present a data-based stochastic simulation method for assessing the efficiency of train timetables for dense traffic areas. We describe models for the random variabilities that occur daily on such networks, and use them in a stochastic simulation algorithm. Results are presented on a saturated line of Paris suburban network.

Keywords: Rail Traffic Management, Theory and Models for Optimization and Control, Simulation and Modeling
Abstract: In the interconnection network operation of plug-in cross-line and post-station crossing, this paper considered the reasonable connection of cross-line train operation, and proposed a method based on coyote optimization algorithm (COA) to solve the train operation adjustment problem in the case of small-scale delay. Firstly, the connection relation of cross-line trains in “Y/T”-type cross-line operation is analyzed. Then, the optimization model is established with the goal of minimizing the total train delay time. Finally, the data of two “Y”-type subway lines in a city are simulated to verify the effectiveness and rapidity of COA in solving the adjustment model, and it is necessary to choose different connections to adjust the delay of different scales. COA achieved better results than IPSO and the computing time was less than 10s, which met the time requirements.

Keywords: Theory and Models for Optimization and Control, Public Transportation Management, Simulation and Modeling
Abstract: Mass transit operators apply operating plans where less frequented stations are not served by all trains, with the aim to shorten the overall passenger travel time. In this paper, we develop two mathematical models in which we explore the possible benefits from two different skipping-stop strategies. In the first model, some stations are skipped by every second train, with the guarantee that every origin-destination pair is feasible without transfer. In the second model, some stations are also skipped by every second train, but without necessarily guaranteeing that every origin-destination pair is feasible without transfer. We adopt here an existing discrete event modeling approach. To have a better overview of the impact of such "skipping stop" policies, we simulate the train dynamics and analyze the skip-stop effects on the service offered to passengers, with our proposed models, and compare the results with those of the case where all trains stop at all stations. The comparison is done in terms of average train frequency and passenger travel time.

Keywords: Public Transportation Management, Rail Traffic Management, Incident Management
Abstract: Pareto dominance is at the core of multi-objective optimization, as it allows different objectives to be simultaneously considered and sets of non-dominated solutions to be built. However in real-time scheduling, determining and using such sets can be challenging. This paper introduces a variation of the commonly used Pareto dominance in the context of real-time rescheduling in a dense railway system. After presenting the industrial context, we describe the academic and industrial motivations in building a new concept of dominance. We then introduce a dominance based on payoff matrices. With this new concept, the size of non-dominated solution sets tends to be reduced and more acceptable solutions are proposed to decision-makers. Preliminary numerical results on industrial data are presented and discussed.

Keywords: Theory and Models for Optimization and Control, Traffic Theory for ITS, Data Mining and Data Analysis
Abstract: This article uses primarily the Grey Relational analysis method to analyze the effectiveness of 14 indicators related to transportation network resilience. In the process of analysis, we use the indicator data obtained from an unattacked network as the optimal reference sequence and a network attacked on the most connected node as the worst reference sequence. Besides the optimal and the worst scenarios, to study the network resilience, we define a network attacking strategy consisting in an assault on one node at a time, orderly for all nodes of the network. A relative Grey Correlation Degree is also proposed to evaluate the results. The analysis is made on 10 public transport networks. They show that the Global Efficiency is the indicator that has the greatest influence on the resilience of the public transportation network. We also categorized the resilience indicators into three different groups. We find that the most important category for network resilience is the Network Efficiency indicator, which includes the network structure plus the bus travel time.

Keywords: Advanced Vehicle Safety Systems, Sensing, Vision, and Perception
Abstract: Ensuring safety for highly automated vehicles (AVs) using complex algorithms including artificial intelligence is still an open research question. A first step towards the goal of achieving safe operation is the Responsibility Sensitive Safety (RSS) model proposed by Intel/Mobileye, which addresses the decision making of an AV system. However, RSS requires a correct environment model. Hence, to have a comprehensive overall AV safety case, additional solutions and argumentation are required, to verify correctness of the environment model. For this purpose we propose in this paper a novel solution, that uses a Monitor-Recovery approach based on a dynamic occupancy grid. The grid is used to verify the object information (Monitor) provided to RSS, and if required to correct wrong information (Recovery). Our results show that with this approach we can detect common failures of a perception system and successfully recover from those errors.

Keywords: Advanced Vehicle Safety Systems, Automated Vehicle Operation, Motion Planning, Navigation, Other Theories, Applications, and Technologies
Abstract: An open question in autonomous driving is how best to use simulation to validate the safety of autonomous vehicles. Existing techniques rely on simulated rollouts, which can be inefficient for finding rare failure events, while other techniques are designed to only discover a single failure. In this work, we present a new safety validation approach that attempts to estimate the distribution over failures of an autonomous policy using approximate dynamic programming. Knowledge of this distribution allows for the efficient discovery of many failure examples. To address the problem of scalability, we decompose complex driving scenarios into subproblems consisting of only the ego vehicle and one other vehicle. These subproblems can be solved with approximate dynamic programming and their solutions are recombined to approximate the solution to the full scenario. We apply our approach to a simple two-vehicle scenario to demonstrate the technique as well as a more complex five-vehicle scenario to demonstrate scalability. In both experiments, we observed an increase in the number of failures discovered compared to baseline approaches.

Keywords: Advanced Vehicle Safety Systems, Automated Vehicle Operation, Motion Planning, Navigation, Other Theories, Applications, and Technologies
Abstract: An open problem for autonomous driving is how to validate the safety of an autonomous vehicle in simulation. Automated testing procedures can find failures of an autonomous system but these failures may be difficult to interpret due to their high dimensionality and may be so unlikely as to not be important. This work describes an approach for finding interpretable failures of an autonomous system. The failures are described by signal temporal logic expressions that can be understood by a human, and are optimized to produce failures that have high likelihood. Our methodology is demonstrated for the safety validation of an autonomous vehicle in the context of an unprotected left turn and a crosswalk with a pedestrian. Compared to a baseline importance sampling approach, our methodology finds more failures with higher likelihood while retaining interpretability.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Advanced Vehicle Safety Systems
Abstract: The safety of Automated Vehicles (AV) as Cyber-Physical Systems (CPS) depends on the safety of their consisting modules (software and hardware) and their rigorous integration. Deep Learning is one of the dominant techniques used for perception, prediction, and decision making in AVs. The accuracy of predictions and decision-making is highly dependant on the tests used for training their underlying deep-learning. In this work, we propose a method for screening and classifying simulation-based driving test data to be used for training and testing controllers. Our method is based on monitoring and falsification techniques, which lead to a systematic automated procedure for generating and selecting qualified test data. We used Responsibility Sensitive Safety (RSS) rules as our qualifier specifications to filter out the random tests that do not satisfy the RSS assumptions. Therefore, the remaining tests cover driving scenarios that the controlled vehicle does not respond safely to its environment. Our framework is distributed with the publicly available staliro and Sim-ATAV tools.

Keywords: Advanced Vehicle Safety Systems, Human Factors in Intelligent Transportation Systems, Driver Assistance Systems
Abstract: Safety is an important challenge in the development of automated vehicles (AVs). To help with the challenge of achieving higher safety in the decision making of AVs, Intel and Mobileye have proposed a parameterized model named Responsibility-Sensitive Safety (RSS). Previous studies have demonstrated that RSS has the potential to improve the safety performance of automated vehicles. However, RSS could lead to a considerable car-following distance depending on the parameter values chosen for the model, which could reduce traffic efficiency. To improve the efficiency of RSS applied to Adaptive Cruise Control (ACC) systems, previous work proposed an efficiency-optimal (referred as “Efficiency-optimal RSS”) variation of the RSS model that involves different triggering conditions of a proper response. Therefore, in this paper a human-in-the-loop driving simulation experiment was conducted to evaluate the performance and acceptance of different safety methods. The RSS model and the efficiency-optimal variant were embedded in an ACC system based on Model Predictive Control (MPC) algorithm. Two car-following scenarios with a sudden deceleration of lead vehicle at various time headways were simulated to evaluate the human perception and response of the different models. Results show that the original RSS model improves subjective safety judgment of human drivers. While the Efficiency-optimal RSS variant has a lower subjective safety score when compared to original RSS, it significantly reduces driver’s emergency braking reactions when compare to an ACC only system.

Keywords: Sensing, Vision, and Perception, Advanced Vehicle Safety Systems, Driver Assistance Systems
Abstract: Driving anomaly detection is an important problem in advanced driver assistance systems (ADAS). The ability to immediately detect potentially hazardous scenarios will prevent accidents by allowing enough time to react. Toward this goal, our previous work proposed an unsupervised driving anomaly detection system using conditional generative adversarial network (GAN), which was built with physiological data and features extracted from the controller area network-Bus (CAN-Bus). The approach generates predictions for the upcoming driving recordings, constrained by the previously observed signals. These predictions were contrasted with actual physiological and CAN-Bus signals by subtracting the corresponding activation outputs from the discriminator. Instead, this study proposes to use a triplet-loss function to contrast the predicted and actual signals. The triplet-loss function creates an unsupervised framework that rewards predictions closer to the actual signals, and penalizes predictions deviating from the expected signals. This approach maximizes the discriminative power of feature embeddings to detect anomalies, leading to measurable improvements over the results observed by our previous approach. The study is implemented and evaluated with recordings from the driving anomaly dataset (DAD), which includes 250 hours of naturalistic data manually annotated with driving events. Objective and subjective metrics validate the benefits of using the proposed triplet-loss function for driving anomaly detection.

Keywords: Sensing, Vision, and Perception
Abstract: Semantic segmentation is an important perception function for automated driving (AD), but training a deep neural network for the task using supervised learning requires expensive manual labeling. Active learning (AL) addresses this challenge by automatically querying and selecting a subset of the dataset to label with the aim to iteratively improve the model performance while minimizing labeling costs. This paper presents a systematic study of deep AL for semantic segmentation and offers three contributions. First, we compare six different state-of-the-art querying methods, including uncertainty-estimate, Bayesian, and out-of-distribution methods. Our comparison uses the state-of-the-art image segmentation architecture DeepLab on the Cityscapes dataset. Our results demonstrate subtle differences between the querying methods, which we analyze and explain. We show that the differences are nevertheless robust by reproducing them on architecture-independent randomly generated data. Second, we propose a novel way to aggregate the output of a query, by counting the number of pixels having acquisition values above a certain threshold. Our method outperforms the standard averaging approach. Finally, we demonstrate that our findings remain consistent for whole images and image crops.