Keywords: .
Abstract:

Keywords: Autonomous Driving
Abstract: Artificial intelligence and machine learning are critical to reaching full autonomy in self driving cars. Two autonomy systems will be presented along with the use of machine learning in each component of them. The presenter will discuss Uber's progress in creating self driving cars for its network and will finish with some observations about the potential impact of these systems in our daily life.

Keywords: Environmental Impact
Abstract: The U.S. Department of Energy’s Vehicle Technologies Office (VTO) supports early-stage research and development of efficient, cost-effective, and sustainable powertrain, vehicle, and transportation technologies that that enable individuals and businesses to save money and use less energy. Through its Energy Efficient Mobility Systems (EEMS) Program, VTO conducts transportation system research at the vehicle, traveler, and system levels, and identifies opportunities to use emerging technologies such as automation and connectivity to improve the mobility of people and goods by making transportation safer, more efficient, and more affordable. The EEMS Program has created sophisticated mobility modeling and simulation tools, developed control algorithms to reduce fuel consumption and improve traffic flow, performed analyses to evaluate the energy and mobility benefits of future transportation scenarios, and studied the important role of traveler decision-making in the transportation system. David Anderson, VTO’s EEMS Program Manager, will discuss why this research area is a priority to the Energy Department, describe major EEMS Program activities, and summarize recent technical results from specific research projects.

Keywords: Collaboration, cooperation, competition, coalitions in traffic and transportation models
Abstract: Technology is pervasive: a good solution to a problem disseminates quickly everywhere, whatever the origin. A recent example is the Internet, which was initiated within the US military and quickly adopted elsewhere including in the USSR. Intelligent Transportation Systems (ITS) has become a global phenomenon, creating smart infrastructures, vehicles and users. On the other hand, surface transportation applications are policy driven and locale oriented, i.e. a transportation system is rooted in a region: Paris is not San Francisco or Shanghai…each often adopts its own policies. What happens with Cooperative ITS (C-ITS)? Isn’t it exacerbated by increasing widely usage of AI? There have been ample examples involving conflicts between innovative technology-based solutions and policies. Transportation authorities often argue the technological solutions do not take enough responsibility. Will these conflicts continue for the deployment of C-ITS? Advancement of robotics and Artificial Intelligence makes it feasible for developers to tune the C-ITS enabled vehicles to meet specific needs and requirements. However, can this lead to locally acceptable systems? How can the existing infrastructure-based policies at the local transportation authority level be reinvented to accommodate the deployment of C-ITS? The International Transport Forum at OECD has begun an interesting global work on data-driven policy. It would be desirable to see globally applicable policies and frameworks, with local parameters to be developed. This talk will give some insights of the dialog between technological innovation communities, the local authorities and the society at large on the challenges of globalization and localization of C-ITS.

Keywords: .
Abstract:

Keywords: Adas, Safety, Intelligent Vehicles
Abstract: Modern ITS (Intelligent Transportation System) applications are becoming increasingly sophisticated and diverse, following the growing trends of connectivity and autonomous driving. For example, a vehicle can be equipped with advanced sensors (e.g., LIDAR, camera or radar) to perform complex driving maneuvers by itself and/or communicate with other vehicles or road-side infrastructures to assist drivers in safer driving. Since the operation of an automotive application is closely tied to the underlying platform characteristics, it requires significant integration effort when the application needs to operate on a wide range of platforms such as other vehicle types or roadside infrastructures. To reduce such effort, we propose a method to implement automotive applications in a platform-independent way. Our approach is to define the automotive domain-specific QoS (Quality of Service) parameters that allow an application to specify the expected quality of sensor input or actuator output independent of a particular platform for safe operation. The application interacts with a platform through several primitive APIs to inform QoS requirements, to read/write sensors/actuators values according to the specified QoS, and to handle exceptions upon any QoS violation. We implement a multi-mode ACC (Adaptive Cruise Control) application on a RC-car platform to demonstrate how an automotive application can be implemented based on the proposed QoS parameters and primitives.

Keywords: Advanced Driver Assistance Systems, Advanced Vehicle Safety Systems, Automated Driving
Abstract: This paper presents a lane departure avoidance system which exploits differential torque delivered to the four wheels as actuation. The system is designed to avoid unsafe lane change due to drowsiness (unintentional maneuver) or inattention (intentional maneuver) and is composed by two levels. A high-level supervisor controls closure of a lateral position control loop by means of a 4-conditions switching rule; a low-level controller is in charge of regulating the vehicle lateral position to the reference value. At first, a model of the system is derived, which describes the dynamics from the control input and steering angle, which acts as a disturbance, to the lateral position. Then, the supervisor is presented and the low-level controller is tuned within the H-infinity framework, with a gain-scheduling term that compensates for speed variations. Finally, the system is shown to be able to stabilize the system without applying excessive control action during typical intentional maneuvers and to keep the vehicle in lane during standard unintentional maneuvers with less than 0.35m error, during straight highway driving.

Keywords: Adaptive Cruise Control
Abstract: We investigate a correct-by-construction synthesis of piecewise affine feedback controllers designed to satisfy the strict safety specifications set forth by the adaptive cruise control (ACC) problem. Our design methodology is based on the formulation of the ACC problem as a reach control problem on a polytope in a 2D state space. The boundaries of this polytope, expressed as linear constraints on the states, arise from the headway and velocity safety requirements imposed by the ACC problem statement. We propose a model for the ACC problem, develop a controller that satisfies the ACC requirements, and produce simulations for the closed-loop system.

Keywords: Connected Vehicles, Automated Driving, Advanced Driver Assistance Systems
Abstract: Conventional lane change warning and automated lane changing systems detect other vehicles using on-board sensors such as camera, radar, and ultrasonic sensors. With the advent of Connected Vehicle (CV) technology, wireless communication (e.g., Dedicated Short Range Communications, or DSRC) becomes another option for “sensing” surrounding vehicles. In particular, DSRC does not have the line-of-sight limitation of ranging sensors and thus can “see” traffic farther ahead, which lends itself well to anticipating the movements of nearby vehicles. We have developed an algorithm that uses such data to predict whether a desired lane change will result in an unsafe situation, and prevents the lane change if that is the case. The effectiveness was evaluated in the microscopic traffic simulator VISSIM using a freeway network that has been well calibrated with rush hour traffic data. System performance in terms of safety was estimated using the Surrogate Safety Assessment Model (SSAM) under a variety of traffic scenarios (different congestion levels, penetration rates of connected vehicles and application-equipped vehicles). Preliminary tests showed that the proposed algorithm can reduce the number of potential traffic conflicts by up to 30%, with higher reductions at higher traffic volumes and higher percentages of application-equipped vehicles.

Keywords: Advanced Driver Assistance Systems, Deep Learning, Driver Classification
Abstract: We propose a driver identification system that uses deep learning technology with controller area network (CAN) data obtained from a vehicle. The data are collected by sensors that are able to obtain the characteristics of drivers. A convolutional neural network (CNN) is used to learn and identify a driver. Various techniques such as CNN 1D, normalization, special section extracting, and post-processing are applied to improve the accuracy of the identification. The experimental results demonstrate that the proposed system achieves an average accuracy of 90% in an experiment with four drivers. In addition, we simulated real-time driver identification in an actual vehicle. In this experiment, we evaluated the time required to reach certain accuracy. For example, the time required to reach an accuracy of 80% was 4–5 min on average.

Keywords: Advanced Driver Assistance Systems, Prediction, Automated Vehicles
Abstract: Situation understanding and interpretation are one of the essential features for automated vehicles. To enable safe and comfortable driving, sensing the current situation is not sufficient, but accurately predicted trajectories of other traffic participants are required. The paper presents a novel trajectory prediction approach utilizing a combination of maneuver classification and probabilistic estimation of temporal properties with a model based trajectory representation. Probabilistic time-to-lane-change estimation is applied to gather information about the conditional distribution for the time of lane marking crossing. Lower tails of the distribution, which represent more critical lane change maneuvers, are utilized with a suitable prediction model to estimate appropriate trajectories. The three parts of the prediction framework are evaluated on the NGSIM data set. It shows, that based on a good performance of the maneuver prediction as well as the time-to-lane-change estimation, lane change trajectories with high accuracy can be predicted. In particular, the consideration of critical maneuver executions shows promising results and demonstrates its general applicability in real-world scenarios.

Keywords: Advanced Driver Assistance Systems, Prediction, Driver Behaviour
Abstract: Automated vehicles require a comprehensive understanding of the current traffic situation and their future evolution to perform safe and comfortable actions. To enable reliable long-term predictions of traffic participants the interaction among each other cannot be neglected. This contribution tackles the problem of interaction-based trajectory prediction with limited information of the situation as delivered by most onboard perception systems of nowadays automated vehicles. A Monte Carlo simulation based approach which utilizes wellknown sensor models in a probabilistic manner to model the interaction between traffic participants is presented. Besides uncertainty in maneuver decisions, the maneuver execution is modeled with a probabilistic model learned from realworld data. Furthermore, the problem of limited information is addressed by a combination of the interaction-aware model with a maneuver classification algorithm providing information on the short-term maneuver. The approach is evaluated on two meaningful simulation scenarios demonstrating the advantages of interaction-based prediction in general and the handling of limited perception capabilities in specific.

Keywords: Computational Intelligence, Agent-based modelling and simulation, Transport Health Management
Abstract: The uneven utilisation of modes of transport has a big impact on traffic in transport pathway infrastrutures. For motor vehicles for instance, this situation explains rapid road deterioration and the large amounts of money invested in maintenance and development due to overuse. There are many approaches to managing this problem; however, the impact of individual users in infrastructural maintenance is mostly ignored. In this position paper, we hypothesise that important changes torwards a more efficient use of the transport network start with its users and their behavioural changes. To this end, we introduce our vision on how to employ data-driven, intelligent agent-based modelling, incorporating human factors aspects, as a toolset to understand travellers and to stimulate behavioural changes. The aim is to achieve better balanced and integrated mobility usage within the transport network. The idea is explored with a few guided questions, and a methodology is proposed. We employ 1) cognitive work analysis to investigate the reasons for travellers’ mode choice; 2) computational intelligence to extract and represent knowledge from related datasets; 3) agent-based modelling to represent the real-world and to observe both individual and emergent behaviours. Future directions to adapt our methodology to alternative smart mobility projects are also discussed.

Keywords: Deep Learning, Transport Health Management, Aviation Systems Intelligent Computational Models
Abstract: The use of sensor technology constantly gathering aircrafts' status data has promoted the rapid development of data-driven solutions in aerospace engineering. These methods assist, for instance, with determining appropriate actions for aircraft maintenance, repair and overhaul (MRO). Challenges however are found when dealing with such large amounts of data. Identifying patterns, anomalies and faults disambiguation, with acceptable levels of accuracy and reliability are examples of complex problems in this area. Experiments using deep learning techniques, however, have demonstrated its usefulness in assisting on the analysis aircraft health data. The purpose of this paper therefore is to conduct a survey on deep learning architectures and their application in aircraft MRO. Although deep learning in general is not yet largely exploited for aircraft health, from our search, we identified four main architectures employed to MRO, namely, Deep Autoencoders, Long Short-Term Memory, Convolutional Neural Networks and Deep Belief Networks. For each architecture, we review their main concepts, the types of problems to which these architectures are employed to, the type of data used and their outcomes. We also discuss how research in this area can be advanced by identifying current research gaps and outlining future research opportunities.

Keywords: Information Security and Privacy, Computational Intelligence, Transport Health Management
Abstract: Vehicular network (VANET), a special type of ad-hoc network, provides communication infrastructure for vehicles and related parties, such as road side units (RSU). Secure communication concerns are becoming more prevalent with the increasing technology usage in transportation systems. One of the major objectives in VANET is maintaining the availability of the system. Distributed Denial of Service (DDoS) attack is one of the most popular attack types aiming at the availability of system. We consider the timely detection and mitigation of DDoS attacks to RSU in Intelligent Transportation Systems (ITS). A novel framework for detecting and mitigating low-rate DDoS attacks in ITS based on nonparametric statistical anomaly detection is proposed. Dealing with low-rate DDoS attacks is challenging since they can bypass traditional data filtering techniques while threatening the RSU availability due to their highly distributed nature. Extensive simulation results are presented for a real road scenario with the help of the SUMO traffic simulation software. The results show that our proposed method significantly outperforms two parametric methods for timely detection based on the Cumulative Sum (CUSUM) test, as well as the traditional data filtering approach in terms of average detection delay and false alarm rate.

Keywords: Traffic Flow Prediction, Off-Line and Online Data Processing Techniques, Prediction
Abstract: Compute efficiency is an important consideration for traffic flow prediction models. Machine learning algorithms adjust model parameters automatically based on the data, but often require users to set additional parameters, known as hyperparameters. Hyperparameters can significantly impact prediction accuracy. Traffic measurements, typically collected online by sensors, are serially correlated. Moreover, the data distribution may change gradually. A typical adaptation strategy is periodically re-tuning the model hyperparameters, at the cost of computational burden. In this work, we present an efficient and principled online hyperparameter learning algorithm for kernel-based traffic prediction models. In tests with real traffic measurement data, our approach requires as little as one-seventh of the computation time of other tuning methods, while achieving better or similar prediction accuracy.

Keywords: Transport Health Management, Data Mining and Data Analysis, Clustering
Abstract: Although a lot of labor and financial forces have been put into safety work, railway accidents continue to be the major concern in China. The aim of this study is to identify the significant factors contributing to railway accidents and enable stakeholders to fully learn from accidents. The Cognitive Reliability and Error Analysis Method - Railway Accidents (CREAM-RAs) taxonomy framework was proposed to classify human, technology, and organization factors in railway accidents. To establish a Multi-attribute Railway Accidents Dataset (MARA-D), 392 railway accident reports were collected and collated under the CREAM-RAs framework. The data mining technique (Self-Organizing Maps – SOM) was adopted to convert MARA-D into 2-dimensional maps. The key accident causes were dug out and risk information was transmitted to various related railway departments. Thus, the relevant measures were raised to improve safety and promote health management of railways.

Keywords: Transport Health Management, Deep Learning, Classification
Abstract: Fault detection in vehicles is currently carried out using model-based or rule-based approaches. Due to advances in automotive technology such as autonomous driving and further connectivity, the complexity of vehicles and their subsystems increases continuously. As a consequence, models and rule-based systems for fault detection become more complex and require more extensive implementation effort and expert knowledge not only within but also across several domains. Besides, vehicles produce rich amounts of data including valuable information for fault detection. These amounts cannot fully be considered by current fault detection approaches. Deep neural networks offer promising capabilities to address these challenges by allowing automated model generation for fault detection without extensive domain knowledge using vast amounts of in-vehicle data. Hence, in this work a data-driven automotive diagnostics approach to fault detection with deep neural networks is proposed in two steps. First, a novel data-driven diagnostics process to learn data-driven algorithms on in-vehicle data is presented. Second, as a key element of this process, a novel fault detection model is proposed using a combination of convolutional and long short-term memory neural networks. To demonstrate the suitability for fault detection, the model is evaluated on internal engine control unit signals for pre-ignition detection in high-pressure turbocharged petrol engines. A classical machine learning processing pipeline and each neural network type separately are used as baselines for a performance comparison. Results show that the proposed model is a promising approach to data-driven automotive diagnostics outperforming all implemented baselines.

Keywords: Transport Health Management, Prediction, Safety
Abstract: It is crucial to predict the remaining useful life (RUL) of aircraft engines accurately and timely for the aircraft operation safety and appropriate maintenance decision. The key issue is how to efficiently mine the internal relation hiding in historical time series monitoring data with high dimension features. In this paper, a data-driven prediction method is proposed by combining the time window (TW) and extreme learning machine (ELM). First, based on the specific properties of aircraft engine time series data, a sliding time window is introduced to sample the historical data to obtain the input vector. Then, the extreme learning machine is utilized to model the relation between time series data and RUL. The proposed approach is validated on the turbofan data sets widely employed by other literatures. Experimental results verify the prediction accuracy and efficiency of the proposed approach compared with the existing methods.

Keywords: Driver Behaviour, Agent-human interactions, Driver Monitoring
Abstract: With increasing automated driving capabilities of commercial vehicles, the study of safe and smooth occupant-vehicle interaction and control transitions is key. In this study, we focus on the development of contextual, semantically meaningful representations of driver and vehicle states, which can then be used to determine the appropriate timing and conditions for transfer of control between driver and vehicle. To this end, we lay out the specifications of the vehicle platform required to conduct such a study, and explore some of the sensors and algorithms that may be needed to produce useful and observable high level cues (features) to make such decisions. These features encode different aspects of the driver state, pertaining to the face, hands, foot and upper body of the driver. Finally, we evaluate these features on their capability of capturing the state of a driver, and demonstrate a strong agreement between these features and a humans' notion of situational awareness.

Keywords: Driver Behaviour, Classification, Artificial Neural Network
Abstract: This paper provides two methods to detect lane changes of the vehicles around the ego vehicle on highway scenarios. The first method uses rule-based with changing lane probabilities over the vehicle distance to the lane and the transverse speed. The second uses neural network with a sliding window of the past trajectory of the vehicle input. Moreover, these methods are compared with a dataset extracted from open road data records with an autonomous vehicle prototype. The results show that the transverse speed as an input of the neural networks has a significant impact on the training and the results and that the rule-based method is slower to detect the lane change than the neural networks.

Keywords: Driver Behaviour, Driver Modelling, Autonomous Vehicles
Abstract: Driving consists of a sequence of interaction with traffic environment and decision making based on situation understanding. Human drivers are capable of taking control in the complex situations smoothly. We thus believe understanding how humans drive and interact with traffic scenes is an important step to achieve an intelligent automated driving system. As the first step to achieve the ultimate goal, in this paper, we presented the Honda Research Institute Driving Dataset (HDD), a challenging dataset to enable research on learning driver behavior and causal reasoning in real-life environments. The dataset includes 104 hours of naturalistic driving collected using an instrumented vehicle. We introduce an annotation scheme to describe complex driving behaviors. A baseline event detection algorithm based on Controller Area Network (CAN bus) data and the corresponding experiments are reported. We also provide a preliminary analysis of the relationship between goal-oriented and stimulus-driven actions.

Keywords: Driver Behaviour, Motion Planning, Intelligent Vehicles
Abstract: In the near future, self-driving vehicles will be frequently tested in urban traffic, and will definitely coexist with human-driving vehicles. To harmoniously share traffic resources, self-driving vehicles need to respect behavioral customs of human drivers. Taking on-road driving for example, self-driving vehicles are supposed to behave in a human-like way to decide when to keep the lane and when to change the lane. This point, however, has not been well addressed in current on-road maneuver decision methods. In this paper, a human-like maneuver decision method based on Long Short Term Memory (LSTM) neural network and Conditional Random Field (CRF) model is proposed for on-road self-driving. Different from previous works, this paper considers the maneuver decision problem as a sequence labeling problem. Its input is a time-series vector which describes a period of neighboring traffic history, and its output is a one-hot vector indicates the suitable maneuver. The proposed model is trained on the NGSIM public dataset, which contains millions of driving maneuvers collected from thousands of human drivers. Simulations with manipulated conditions reveal human-like reasoning for maneuver decision inside the proposed model. Comparative experiments further demonstrate a better human-like performance achieved by the proposed method than that of previous methods.

Keywords: Driver Behaviour, Prediction, Public Transport
Abstract: Santiago’s public transit system uses a Passenger Service Quality Index (ICA) to measure the quality of service offered by buses companies. Parts of this index are related to bus driver’s behavior, and are obtained in a superficial and very subjective manner. The main objective of this research is to formulate a new methodology that uses data provided by inertial measurement units to classify drivers’ behavior. This is achieved by means of a classification method: decision trees. Data are collected to evaluate the method and results show that the use of decision trees delivers a good performance. Also, the model delivers an interpretable output that allows further analysis. The classification made is analyzed and a consistency with the current methodology is found.

Keywords: Human Factors, Agent-human interactions, Autonomous Vehicles
Abstract: The comfort level of passengers is an important factor in measuring user experience in any form of transportation, including in autonomous vehicles. One of the main factors that determines user acceptance of autonomous vehicles is the passenger’s level of discomfort in a ‘control - and authority-less’ experience. In this paper, we propose an approach for formulating discomfort through ‘on-the-road’ field studies, with human driven vehicles, while the passenger provides real-time explicit feedback on discomfort via a potentiometer. While previous studies focused on the association between vehicle dynamics and passenger discomfort, we demonstrate here how we can improve the classification ability of passenger discomfort by employing a multi-dimensional model that also takes into account the external scenario (contextual information). This is achieved by processing image data (e.g. distance from nearest bicycle) recorded through an outward looking camera in addition to location/route data obtained from other sensors like GPS. As such, the focus of this paper is on classification of external information.

Keywords: Human Factors, Driver State, Advanced Driver Assistance Systems
Abstract: Advanced driver assistance systems (ADAS) for cars have been in market for a few decades and gaining popularity. The automation level for these systems are getting higher over years and automated driving is expected to be launched in near future. Advantage of those systems is not only for safety, but also for reducing the workload in driving. Especially the system which allows drivers to leave their hands off the steering wheel is considered to provide additional benefits to drivers compared to the system requiring hands on the steering wheel or to manual driving. The one of the additional benefits is the mental workload reduction, in other words stress level reduction, by free of their hands in driving. In this paper, we propose the method to measure mental workload which allows quantitative comparison in stress level between hands off and hands on the steering wheel system including manual driving, taking individual initial stress level and temporal change of stress level in a day into account. The proposed method is relatively easier on the measurement procedure and not requires complex measurement tools. In this sense, it fits to evaluate ADAS systems with actual driving. We report with experiments that our proposed method is effective for the purpose of evaluating the mental workload reduction for highly advanced driver assistance systems.

Keywords: Data Mining and Data Analysis
Abstract: With the aggravation of traffic congestion, lots of cities in China have begun to carry out the license plate restriction policies based on license number to relieve traffic congestion. Based on the vehicles’ license plate identification data from cameras in Hangzhou, the effect of restriction policy especially during peak hours on traffic conditions, travel time, and number of trips have been evaluated. Also, commuter vehicles were identified considering their departure or arrival time, time of day, and travel frequency. The trip behavior of restricted commuter vehicles was studied. Based on the obtained results, the research offers some proposals about Hangzhou’s restriction policies. Moreover, the research will also provide basis for conducting and changing the restriction policies.

Keywords: Ride Sharing, Multi-modality, Data Mining and Data Analysis
Abstract: Both of the transportation and the public hospital outpatient services are the key public services provided by the government, which guarantee the safe, healthy and worthwhile life for daily living in metropolitan cities such as Beijing and Hong Kong. Since the citizens usually have to see their medical doctors at the appointed time, they should arrive punctually as scheduled for their outpatient services in public hospitals. In order to find out how people travel to the public hospitals for their scheduled outpatient appointments particularly before and after the bike-sharing system in Beijing, two interview surveys with a total of 1,047 valid samples were conducted at the Beijing Friendship Hospital in October 2016 and 2017. One was before the launch of the floating bike-sharing system in November 2016 while the other one was conducted in October 2017 after this new system has been implemented in Beijing for about one year. Compared to the other non-motorized travel modes, the bike-sharing mode could provide the best trip experience with the highest evaluation scores on the satisfaction, the comfort, the convenience and the punctuality of the outpatient trips. Multinomial Logit (MNL) model is calibrated to investigate the factors to influence the travel mode choice for outpatient trips. The results show that the outpatient travelers are likely to choose the bike-sharing mode for their return outpatient visits at public hospitals in Beijing. Other factors like online map-usage, the acceptable maximum walking time and cycling time, the weekend appointment for outpatient services and the household income have positive impacts on the choice of the bike-sharing mode.

Keywords: CNN, Detection, Environment Perception
Abstract: Traffic lights classification is a very important task that should be accomplish using computer vision techniques. RADAR or LIDAR sensors are suitable to detect traffic lights. However, they are not able to distinguish between traffic light states. This critical task embraces passengers safety during autonomous driving and it can only be solved using computer vision approaches. In this paper, a wide analysis of the state of the art regarding traffic lights classification is performed. The proposed approach is based on a ResNet architecture and it is compared against a more complex architecture (MobilNet) and also against a traditional feature-based classifier (Random Trees). Due to the importance of high quality training data, a comparative analyisis of the existing datasets related with traffic light classification is presented. The analysis take into account image patch size, number of labels and number of samples for each label.

Keywords: Detection, Convolutional Neural Networks, Autonomous Driving
Abstract: Reliable traffic light detection is one crucial key component for autonomous driving in urban areas. This includes the extraction of direction arrows contained within the traffic lights as an autonomous car will need this information for selecting the traffic light corresponding to its current lane. Current state of the art traffic light detection systems are not able to provide such information.

Within this work we present a hierarchical traffic light detection algorithm, which is able to detect traffic lights and determine their state and contained direction information within one CNN forward pass. This Hierarchical DeepTLR (HDTLR) outperforms current state of the art traffic light detection algorithms in state aware detection and can detect traffic lights with direction information down to a size of 4 pixel in width at a frequency of 12 frames per second.

Keywords: Human-Machine Interface, Traffic Flow, Driver Assistance Systems
Abstract: The efficiency of traffic flow in urban areas can be improved with a Traffic Light Assistant (TLA), which indicates the status of upcoming traffic lights based on a driver’s current traveling speed. Additionally, TLAs can help reduce the number of stops at traffic lights, which will also have a positive effect on fuel consumption. In the current paper, two different Human Machine Interfaces (HMIs) for a TLA were designed as smartphone applications with multimodal interactions. Driver glance behavior was tested according to the NHTSA guideline. The results show the outcomes of a suitability study carried out in a static driving simulator. Both HMI concepts (Perspective HMI: 915 ms; Two-Dimensional HMI: 849 ms) fulfill the NHTSA requirement that the 85th percentile of single glance duration is to less than 2 seconds.

Keywords: Detection, Camera, Convolutional Neural Networks
Abstract: Recent improvements in object detection are driven by the success of convolutional neural networks (CNN). They are able to learn rich features outperforming hand-crafted features. So far, research in traffic light detection mainly focused on hand-crafted features, such as color, shape or brightness of the traffic light bulb. This paper presents a deep learning approach for accurate traffic light detection in adapting a single shot detection (SSD) approach. SSD performs object proposals creation and classification using a single CNN. The original SSD struggles in detecting very small objects, which is essential for traffic light detection. By our adaptations it is possible to detect objects much smaller than ten pixels without increasing the input image size. We present an extensive evaluation on the DriveU Traffic Light Dataset (DTLD). We reach both, high accuracy and low false positive rates. The trained model is real-time capable with ten frames per second on a Nvidia Titan Xp.

Keywords: Computer Vision, Advanced Driver Assistance Systems, Recognition
Abstract: We propose a novel, real-time framework for traffic sign detection and recognition using a camera mounted on the dashboard of a moving vehicle. Traffic sign detection and recog- nition plays an important role in Advanced Driver Assistance Systems (ADAS) as it helps increase driving safety. However, it is very challenging to detect and recognize traffic signs because of challenges such as perspective distortion, illumination variation, occlusion and motion blur from moving vehicle.

In our framework, we use Hue Saturation Value (HSV) color filtering for traffic sign detection. In our Grassmann manifold based traffic sign recognition framework, we create subspaces of each unique traffic sign. These subspaces accommodate the uncertainties that occur during detection and variations in different instances of the same traffic sign. These subspaces lie on a Grassmann manifold and we use discriminant analysis on Grassmann manifolds for recognising them. We have carried out extensive experiments on multiple publicly available traffic sign datasets and compared our proposed framework with multiple state-of-the-art methods. Experimental results show that our system is robust and has a high degree of accuracy.

Keywords: Sensing, Vision, and Perception, Autonomous Driving, Deep Learning
Abstract: Traffic lights perception problem is one of the key challenges for autonomous vehicle controllers in urban areas. While a number of approaches for traffic light detection have been proposed, these methods often require a prior knowledge of map and/or show high false positive rates. Recent successes suggest that deep neural networks will be widely used in self-driving cars, but current public datasets do not provide sufficient amount of labels for training such large deep neural networks. In this paper, we developed a two-step computational method that can detect traffic lights from images in a real-time manner. The first step exploits a deep neural object detection architecture to fine true traffic light candidates. In the second step, a point-based reward system is used to eliminate false traffic lights out of the candidates. To evaluate the proposed approach, we collected a human-annotated large-scale traffic lights dataset (over 60 hours). We also designed a real-world experiment with an instrumented self-driving vehicle and observed that the proposed method was able to handle false traffic lights substantially better compared with the baseline considered.

Keywords: Deep Reinforcement Learning, Reinforcement Learning, Traffic Signal Control
Abstract: The population in cities and demand for transportation continuously increases. Space, financial and environmental constraints do not allow for significant infrastructure expansion. Therefore, optimizing the efficiency of the infrastructure is becoming increasingly important. Wait time at traffic lights is a significant proportion of time spent travelling within cities. Time inefficiency of traffic lights is, therefore, a global concern. Adaptive traffic signal controllers aim to provide demand-responsive strategies to minimize motorists’ delay and achieve higher throughputs at signalized intersections. With the advent of new sensory technologies and more intelligent control methods, we propose an adaptive traffic signal controller able to receive un-prepossessed high-dimensional sensory information and self-learn to minimize the intersection delay. We use (1) deep neural networks to operate directly on detailed sensory inputs and feed it into (2) a continuous reinforcement learning based optimal control agent. The integration of the two is known as deep reinforcement learning or deep learning for short. Using deep learning, we achieve two goals: (1) eliminate the need for handcrafting a feature extraction process such as determining queue lengths for instance, which is challenging and location specific, and (2) achieve better performance and faster training time compared to conventional discrete reinforcement learning approaches.

Keywords: Energy Storage and Control Systems, Electric Vehicles, Electric Motors, Drives and Propulsion Technologies
Abstract: In this paper, we introduce the combination of thermostat control strategy (TCS) and DC-link voltage control strategy to improve fuel economy for series hybrid electric vehicles (HEVs). The main method is tuning essential parameters, including a parameter under DC-link voltage control and a selected constant as an important judgement condition of charging operation modes under TCS. The minimum mass points of equivalent fuel consumption (EFC) corresponding to a series of variables are marked for worldwide harmonized light vehicles test procedure (WLTP). By analysis and verification, the fuel economy of series HEVs with WLTP performs better with the combination control schemes than individual control scheme, so this scheme proves very effective for series HEVs driving in an urban environment.

Keywords: Electric Autonomous Vehicles, Automated Vehicle Operation, Motion Planning, Navigation, Vehicle Safety
Abstract: In this paper is presented a linear MPC controller design for autonomous cars navigation. It combines both the lateral and longitudinal control. The MPC cost function has been designed to account for human driving behaviours, i.e. it smoothes out coarse reference trajectories. Furthermore, a safety monitoring module has been implemented. It computes an estimated time before reaching an unacceptable situation (w.r.t. comfort constraints and tracking performance) under the current tracking conditions. The overall benefit of this controller is to guarantee trajectory smoothness while outputting information on its performance. This information will later be used to re-plan safe trajectories in dynamic environments. The proposed linear MPC controller has been tested in a typical urban scenario.

Keywords: Electric Autonomous Vehicles, Artificial Intelligence, Deep Q-Network
Abstract: The dependence of autonomous vehicles (AVs) on sensors and communication links exposes them to cyber-physical (CP) attacks by adversaries that seek to take control of the AVs by manipulating their data. In this paper, the state estimation process for monitoring AV dynamics, in presence of CP attacks, is analyzed and a novel adversarial deep reinforcement learning (RL) algorithm is proposed to maximize the robustness of AV dynamics control to CP attacks. The attacker's action and the AV's reaction to CP attacks are studied in a game-theoretic framework. In the formulated game, the attacker seeks to inject faulty data to AV sensor readings so as to manipulate the inter-vehicle optimal safe spacing and potentially increase the risk of AV accidents or reduce the vehicle flow on the roads. Meanwhile, the AV, acting as a defender, seeks to minimize the deviations of spacing so as to ensure robustness to the attacker's actions. Since the AV has no information about the attacker's action and due to the infinite possibilities for data value manipulations, each player uses long short term memory (LSTM) blocks to learn the expected spacing deviation resulting from its own action and feeds this deviation to a reinforcement learning (RL) algorithm. Then, the attacker's RL algorithm chooses the action which maximizes the spacing deviation, while the AV's RL algorithm seeks to find the optimal action that minimizes such deviation. Simulation results show that the proposed adversarial deep RL algorithm can improve the robustness of the AV dynamics control as it minimizes the intra-AV spacing deviation.

Keywords: Electric Autonomous Vehicles, Electric Motors, Drives and Propulsion Technologies, Vehicle Safety
Abstract: To enhance the reliability of the powertrain system of automated vehicles, a diagnostic concept to detect various sensor faults is presented in this paper. By analyzing the signals of existing sensors in the vehicle and using a physical model of a powertrain system, the analytic redundancy relations between signals are derived. Based on the structural system analysis theory, eleven residuals are designed in this paper to detect and isolate various sensor faults in the powertrain system, e.g. voltage, current and position sensors. The proposed method has been implemented and simulated in Matlab/Simulink. As a result, the analyzed sensor faults can be detected successfully. This new method shows good robustness against measurement tolerance and model uncertainty.

Keywords: Energy Storage and Control Systems, Electric Vehicles, Prediction
Abstract: Among the biggest challenges battery electric vehicles (BEV) face is their limited range and higher price compared to conventionally powered vehicles. One underlying reason is the operation strategy of the battery pack. To reduce aging and avoid deep discharge, state of the art battery management employs strict and conservative operational limits, leaving a significant amount of energy in the battery unused, directly translating into a loss of possible range. Additionally, this strategy leads to a sudden and, for the driver unpredictable, power derating at a low state-of-charge (SOC).

This paper proposes an approach for defining the operational limits of the vehicle's battery and a strategy for using most of the available capacity and power at a low (SOC), increasing the vehicle's capabilities with no additional hardware. The benefits of this approach are shown by a theoretical assessment of selected (BEV). The limitations of current implementations are analyzed by an experimental study of a Volkswagen e-Golf on a chassis dynamometer and under real driving conditions.

The impact of an extension of the battery's operational limits is discussed with a study of relevant literature and supported by measurements of lithium-ion battery behavior. The strategies presented in this paper are implemented, simulated and evaluated under consideration of the behavior of a single cell within the battery pack, as well as the whole electric drivetrain.

Keywords: Planning, Electric Vehicles, Smart Mobility
Abstract: A major challenge for large-scale deployment of electrical vehicles (EVs) is charging. In general, the number of EVs that can be charged can be increased either by physically expanding charging capacity or by better exploiting existing charging capacity. In this paper, we focus on the latter, exploring how advanced EV travel planning systems can be used to better align where and when EV charging happens with where and when charging capacity is available. Our novel travel planner enables EV users to plan their trips and EV charging in a way that meets their needs yet reflects charging availability. The core innovation of our approach is that we take a broader, multi-destination perspective to EV travel planning -- this gives our planning system more flexibility and scope for deciding when and where charging should happen and, consequently, enables a better alignment between the need for charging implied by the EV travel and the availability of charging. We evaluate our approach on an agent-based simulation of several medium-scale scenarios based on real-world data. The results confirm the benefits of our multi-destination approach, especially in scenarios in which charging service providers support upfront booking of charging slots.

Keywords: Traffic Control, Cooperative Driving, Multi-Autonomous Vehicle Studies, Models, Techniques and Simulations
Abstract: Autonomous vehicles are currently being tested in a variety of scenarios. As we move towards autonomous vehicles, How do Intersections need to look? To answer that question, We break down an intersection management into the different conundrums and scenarios involved in the trajectory planning and current approaches to solve them. Then a brief analysis of current works in autonomous intersection is performed. With a critical eye, we try to delve into the current solutions' discrepancies while providing some critical and important factors that have been addressed and look at open issues that have to be addressed. We also try to answer the question of how to benchmark intersection management algorithms by providing some factors that have an impact on the system.

Keywords: Traffic Signal Control, Automated Vehicles, Connected Vehicles
Abstract: Connected and automated vehicle (CAV) technology has progressed to the point where these vehicles are beginning to appear in traffic streams around the globe. As a result, there is a rising need for similar advances in the technology underlying traffic signal control systems. In this paper, we perform a quantitative and qualitative comparison of two state-of-the-art intelligent signal control systems; namely, the Hybrid Autonomous Intersection Management (H-AIM) cite{sharon2017intersection} system and the Intelligent Intersection Control Algorithm (IICA) cite{PourmehrabERM17}. These systems use real-time sensors such as traffic cameras, vehicle-to-infrastructure communication, and real-time optimization and/or decision-making to maximize the efficiency of signalized intersections for CAVs as well as conventional vehicles. We describe the various assumptions under which these systems operate and present simulation results over a range of traffic scenarios and CAV penetration rates. Under the considered headways, IICA achieves the lowest travel times and highest throughput for CAV penetration rates up to around 95%.

Keywords: Advanced Driver Assistance Systems, Cooperative Automated Vehicles, Motion Planning
Abstract: This research proposes a cooperative lane-changing (CLC) controller for connected and automated vehicles under partially connected and automated environment. The goal is to further advance the technology of automated lane change in terms of realizing coordination and cooperation between a lane change vehicle and its following vehicle in order to reduce oscillation and shockwave caused by lane change maneuvers. The proposed controller is formulated as a model predictive control. The optimal control problem is solved by a dynamic programming based numerical algorithm proposed by this research team in a previous study to ensure computation speed. The proposed CLC controller is evaluated against human drivers to quantify its performance. Sensitivity analysis is conducted in terms of the initial headway of receiving gap. The results proved the effectiveness of the CLC controller. For a lane-change vehicle, CLC reduced oscillation by 0.5%-13.9%. For its following vehicle, CLC benefited up to 9.6%. The variation is due to the changes in initial headway of receiving gap. The computation time is around 17-21 ms with 0.1 seconds time step and 20 seconds optimization time horizon.

Keywords: Electric Motors, Drives and Propulsion Technologies, Shortest Path, Advanced Driver Assistance Systems
Abstract: The accuracy of the modern navigation and traffic information systems has increased noticeably. This offers the opportunity of improving the standard powertrain energy management of hybrid electric vehicles (HEVs) by making use of predictive information about the upcoming route. Eco-routing for HEVs aims to optimize fuel consumption by deciding the best route and when to use or recover electrical power based on topological and traffic information. In this work, a simple predictive speed model is used to derive a fast semi-analytical solution of the powertrain energy management. The predicted fuel consumption on each road segment is a function of the desired trade-off between fuel economy and battery use. This modeling complexity is addressed by introducing a novel road network model which takes into account the feasible battery charge variation at road segment level. Also, the routing of HEVs adds several intrinsic constraints and complexity to the classical shortest path problem (SPP). A relaxation of the constrained SPP is proposed to reduce computation time and thus increase user acceptance.

Keywords: Advanced Driver Assistance Systems, Driver Assistance Systems, Driver Modelling
Abstract: Vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications have the potential to not only accelerate progress towards autonomous vehicles but also to facilitate the achievement of more appropriate driver behavior patterns and thus make traffic more energy efficient and environmentally friendly. This paper presents a driving advisory system for vehicles approaching traffic lights that is based on the use of a Green Light Optimal Speed Advisory (GLOSA) system and evaluates its impact on traffic flow characteristics. This system, which assumes the existence of a connection with the traffic light ahead, calculates the distance required to pass through the intersection without stopping during the remaining green signal time. The required distance is then projected onto the road surface as a green planar rectangle shape through a head-up display (HUD). As long as the vehicle is traveling on the green rectangle, the drivers will not need to make a complete stop or accelerate again in order to meet the onset of a green signal. Assuming that all vehicles are equipped with our proposed system, traffic simulations were carried out to evaluate the performance of the scheme. Numerical analyses showed that a higher fuel efficiency and lower carbon dioxide (CO2) emissions were achievable, even under high-density traffic conditions, by combining the full and partial assistance levels of our proposed system.

Keywords: Cooperative Adaptive Cruise Control, Traffic Simulation, Travel Time
Abstract: With the rapid development and wide deployment of intelligent vehicles, they will exert significant impacts on traffic efficiency not only in developed countries, but also in developing countries such as China. This paper focuses on the effects of Cooperative Adaptive Cruise Control and Intelligent Speed Adaption on Chinese urban, highway and rural traffic. Through several parallel microscopic traffic simulations of Chinese contexts considering different road types and penetration rates, we find that these intelligent vehicle technologies can generate noticeable decrease in travel time and increase in travel speed for urban traffic, especially in the morning and evening peak time, but not quite effective on highways and in rural areas. The study also indicates the importance of their market penetration rate to technological effects. Furthermore, based on the characteristics and simulation results of Chinese traffic, we propose some suggestions for developing countries to make better use of these advanced technologies.

Keywords: Urban Traffic Control, V2v Communication, Cooperative Adaptive Cruise Control
Abstract: With the emergence of autonomous ground vehicles and the recent advancements in Intelligent Transportation Systems, Autonomous Traffic Management has garnered more and more attention. Autonomous Intersection Management (AIM), also known as Cooperative Intersection Management (CIM) is among the more challenging traffic problems that poses important questions related to safety and optimization in terms of delays, fuel consumption, emissions and reliability. Previously we introduced two stop-sign based policies for autonomous intersection management that were compatible with platoons of autonomous vehicles. These policies outperformed regular stop-sign policy both in terms of average delay per vehicle and variance in delay. This paper introduces a reservation-based policy that utilizes the cost functions from our previous work to derive optimal schedules for platoons of vehicles. The proposed policy guarantees safety by not allowing vehicles with conflicting turning movement to be in the conflict zone at the same time. Moreover, a greedy algorithm is designed to search through all possible schedules to pick the best that minimizes a cost function based on a trade-off between total delay and variance in delay. A simulator software is designed to compare the results of the proposed policy in terms of average delay per vehicle and variance in delay with that of a 6-phase

Keywords: Artificial Neural Network, Convolutional Neural Networks, Classification
Abstract: When parameterizing vehicle control algorithms for stability or trajectory control, the road-tire friction coefficient is an essential model parameter when it comes to control performance. One major impact on the friction coefficient is the condition of the road surface. A camera-based, forward-looking classification of the road-surface helps enabling an early parametrization of vehicle control algorithms. In this paper, we train and compare two different Deep Convolutional Neural Network models, regarding their application for road friction estimation and describe the challenges for training the classifier in terms of available training data and the construction of suitable datasets.

Keywords: Road Edge Detection, Autonomous Vehicles, Segmentation
Abstract: The classification of semantically meaningful road markings in images is an important and safety critical task for autonomous and semi-autonomous vehicles. However, beyond simple lane markings, real-time detection and interpretation of road markings is challenging as images are subject to occlusions, partial observations, lighting changes and differing weather conditions. Additionally, there is high variation in the road markings between countries and regions, which makes interpretation difficult. In this work we present a three-fold approach to the semantic classification. Firstly, we employ a weakly supervised neural network to detect pixels belonging to road markings under different conditions. Subsequently, these pixels are classified into geometric primitives, from which we retrieve the semantic classes through a fast and parallel model-fitting algorithm that offers real-time performance. Unlike other methods in the literature that perform road marking classification independently, our proposed approach performs a joint classification leveraging the highly structured configurations that characterise urban traffic scenes. Consequently, we retrieve the underlying semantic classes under a variety of weather and lighting conditions as we demonstrate in our results.

Keywords: Computer Vision, Road Edge Detection, Crowdsourcing
Abstract: This paper proposes an algorithm for camera based roadway mapping in urban areas. With a convolutional neural network the roadway is detected in images taken by a camera mounted in the vehicle. The detected roadway masks from all images of one driving session are combined according to their corresponding GPS position to create a probabilistic grid map of the roadway. Finally, maps from several driving sessions are merged by a feature matching algorithm to compensate for errors in the roadway detection and localization inaccuracies. Hence, this approach utilizes solely low-cost sensors common in usual production vehicles and can generate highly detailed roadway maps from crowd-sourced data.

Keywords: Scene Parsing, Artificial Intelligence, Perception
Abstract: Autonomous vehicles require an accurate and adequate representation of their environment for decision making and planning in real-world driving scenarios. While deep learning methods have come a long way providing accurate semantic segmentation of scenes, they are still limited to pixel-wise outputs and do not naturally support high-level reasoning and planning methods that are required for complex road manoeuvres. In contrast, we introduce a hierarchical, graph-based representation, called scene graph, which is reconstructed from a partial, pixel-wise segmentation of an image, and which can be linked to domain knowledge and AI reasoning techniques. In this work, we use an adapted version of the Earley parser and a learnt probabilistic grammar to generate scene graphs from a set of segmented entities. Scene graphs model the structure of the road using an abstract, logical representation which allows us to link them with background knowledge. As a proof-of-concept we demonstrate how parts of a parsed scene can be inferred and classified beyond labelled examples by using domain knowledge specified in the Highway Code. By generating an interpretable representation of road scenes and linking it to background knowledge, we believe that this approach provides a vital step towards explainable and auditable models for planning and decision making in the context of autonomous driving.

Keywords: Road Edge Detection, Perception, Deep Learning
Abstract: This paper is about the detection and inference of road boundaries from mono-images. Our goal is to trace out, in an image, the projection of road boundaries irrespective of whether or not the boundary is actually visible. Large scale occlusion by vehicles prohibits direct approaches - many scenes present 100% occlusion and so we must infer the boundary location using scene context. Such a problem is well suited to CNN derived approaches but the sinuous structure of a hidden narrow continuous curve running through the image presents challenges for conventional NN-architectures. We approach this as a coupled, two class detection problem -solving for occluded and non-occluded curve partitions with a continuity constraint. Our network output is in a hybrid discrete-continuous form which we interpret as measurements of segments of the true road boundary. These measurements are passed to a model selection stage which associates measurements to minimal number of a-priori unknown set of geometric primitives (cubic curves) representing road boundaries. We present a semi-supervised method which leverages a visual localisation to generate 25 thousand labelled images for training and testing - the results of which are presented in the conclusion of the paper.

Keywords: Road Edge Detection, Environment Perception, Autonomous Driving
Abstract: In this paper, we studied a SLAM method for vector-based road structure mapping using multi-beam LiDAR. We proposed to use the polyline as the primary mapping element instead of grid cell or point cloud, because the vector-based representation is precise and lightweight, and it can directly generate vector-based High-Definition (HD) driving map as demanded by autonomous driving systems. We explored: 1) the extraction and vectorization of road structures based on local probabilistic fusion. 2) the efficient vector-based matching between frames of road structures. 3) the loop closure and optimization based on the pose-graph. In this study, we took a specific road structure, the road boundary, as an example. We applied the proposed matching method in three different scenes and achieved the average absolute matching error of 0.07 m. We further applied the vector-based mapping system to a road with the length of 860 meters and achieved an average global accuracy of 0.466 m without the aid of GPS.

Keywords: Road Edge Detection, Computer Vision, Camera
Abstract: In this paper, a lane detection and classification algorithm based on a monocular camera for a two-wheeled vehicle in small roll angle conditions is described. The algorithm is designed to work in different illumination conditions, both in night time and daytime. First, all road line markings are identified using a linear lane model exploiting Hough transform and perspective filtering. Then, identified line markings are classified based on geometric features using a suitable SVM. Then, the boundaries of the actual lane occupied by the vehicle are detected among all the possible lane marking couples. Finally, detected lane is tracked using heuristic rules. The overall strategy has been validated using videos acquired with a motorcycle in different scenarios. Experimental results have proven the robustness of the algorithm with respect to illumination changes and small variations of motorcycle roll angle, with a overall detection accuracy of 94.8% and classification accuracy of 98.1%.

Keywords: Autonomous Driving, Data Mining and Data Analysis, Artificial Intelligence
Abstract: Many modern approaches for autonomous vehicles are still limited to a low-level data representation without considering complex and relational information about their environment. This often leads to a generalization problem in complex situations where algorithms only perform well in tailored scenes. The general challenge lies in combining probabilistic information about partially observable traffic participants with a semantic description of the environment to reduce the complexity of a scene making it manageable. In this work, we present a novel approach for considering high-level information in traffic situations with respect to modeling a semantic human-readable representation of the vehicles’ environment as well as uncertainties respecting their probabilities through relations between entities. The knowledge is described by first-order logic in a formal language. The environmental information is stored in a multidimensional tensor which we call Semantic Tensor. It allows to model complex scenes and functions as a very fast data structure. A novel, general similarity measure allows comparing any two Semantic Tensors to extract information about how similar two thus represented situations are. We show its applicability on a complex scenario which is not fully observable.

Keywords: Cooperation, Vulnerable traffic users, Data Association, Extend Kalman Filter
Abstract: In future traffic scenarios, vehicles and other traffic participants will be interconnected and equipped with various types of sensors, allowing for cooperation based on data or information exchange. This article presents an approach to cooperative tracking of cyclists using smart devices and infrastructure-based sensors. A smart device is carried by the cyclists and an intersection is equipped with a wide angle stereo camera system. Two tracking models are presented and compared. The first model is based on the stereo camera system detections only, whereas the second model cooperatively combines the camera based detections with velocity and yaw rate data provided by the smart device. Our aim is to overcome limitations of tracking approaches based on single data sources. We show in numerical evaluations on scenes where cyclists are starting or turning right that the cooperation leads to an improvement in both the ability to keep track of a cyclist and the accuracy of the track particularly when it comes to occlusions in the visual system. We, therefore, contribute to the safety of vulnerable road users in future traffic.

Keywords: Cooperative Automated Vehicles, Collaboration, cooperation, competition, coalitions in traffic and transportation models, Motion Planning
Abstract: Tactical maneuver planning of multiple, communicating vehicles provides the opportunity to increase passenger safety and comfort. We propose a unifying method to orchestrate the motion of cooperative vehicles based on the negotiation of conflicting road areas, which are determined by reachable set computation. As a result, each vehicle receives an individual driving corridor for trajectory planning. The presented conflict resolution scheme has polynomial runtime complexity and is guaranteed to find the optimal allocation of road areas for each negotiation round. Our method is not tailored to specific traffic situations but is applicable to general traffic scenes with manually driven and automated vehicles. We demonstrate the universal usability of our approach in numerical experiments.

Keywords: Cooperative Automated Vehicles, Cooperative Driving, Trajectory Planning
Abstract: Urban traffic scenarios often require a high degree of cooperation between traffic participants to ensure safety and efficiency. Observing the behavior of others, humans infer whether or not others are cooperating. This work aims to extend the capabilities of automated vehicles, enabling them to cooperate implicitly in heterogeneous environments. Continuous actions allow for arbitrary trajectories and hence are applicable to a much wider class of problems than existing cooperative approaches with discrete action spaces. Based on cooperative modeling of other agents, Monte Carlo Tree Search (MCTS) in conjunction with Decoupled-UCT evaluates the action-values of each agent in a cooperative and decentralized way, respecting the interdependence of actions among traffic participants. The extension to continuous action spaces is addressed by incorporating novel MCTS-specific enhancements for efficient search space exploration. The proposed algorithm is evaluated under different scenarios, showing that the algorithm is able to achieve effective cooperative planning and generate solutions egocentric planning fails to identify.

Keywords: Agent-based modelling and simulation, Microscopic Traffic Simulation, Electric Vehicles
Abstract: Electric vehicles (EVs) are expected to play major roles in the pursuit of a low-carbon society; hence, using renewable energy to effectively charge EVs will be crucial. Interactions between road transport networks, in which gasoline-powered vehicles are used, and electric power systems maintained by controllable energy systems have not been considered. However, EVs, which are becoming increasingly popular, will spur such interactions. In this study, we propose a coupled simulation model that can represent the interactions between transport and electric power systems. Assuming high-output charging using fast chargers at charging stations for long-distance trips and low-output charging using a normal charger at home, we evaluate the time series change of load flow in urban power systems resulting from EV charging events.

Keywords: Artificial Intelligence, Data Mining and Data Analysis, Artificial Neural Network
Abstract: It’s getting more difficult to find available parking spaces with the rapid growth of the vehicles, especially in big cities. The parking guidance system (PGS) which powered by real time data or historical data could reduce the time spent on looking for parking spaces and alleviate the heavy traffic around the parking lots. However, the PGS is completely ineffective when the parking lots missing data or even have no parking data. This paper makes an exploration on data repairing by digging geospatial data and historical data. First, this paper proposes a method to verify the possibility of parking data similarity when parking lots have spatial similarity. Then take the known data of parking lots as samples and generate reparative data through the Recurrent GANs. The experiment shows, when parking lots have similar spatial features, the parking data, in a high probability, have some similarity with each other, too. And the data generated by Recurrent GANs have the same distribution with real data. In fact, this paper provides a new idea to solve the problems of timeseries data repairing, to a certain extent, this method can help to save the cost of equipment and time.

Keywords: Communications and Protocols in ITS, Internet of Things, V2v Communication
Abstract: Recently low power wide area network (LPWAN) is widely researched and deployed due to its excellent performance of supporting large coverage, low power consumption and massive capacity. LPWAN might offer brandnew solutions to Vehicle to anything (V2X) communications, which is faced with the challenge of supporting massive connections due to the dramatically increasing number of vehicles. In this paper, after surveying the existing V2X communication technologies, we compare the traditional technologies with the representative LPWAN technologies according to their performance metrics. After the careful comparison and selection, Long-Range (LoRa) and enhanced machine type communication (eMTC) are introduced to V2X communication due to their support for mobility. Moreover, their performance are evaluated in both V2I (vehicle-to-infrastructure) and V2V (vehicle-to-vehicle) communication environments via Monte Carlo simulations.

Keywords: Growing artificial societies in artificial transportation systems, Agent-based modelling and simulation, Large scale simulation of agent-based traffic models
Abstract: Synthetic population is one of the most important foundations of disaggregated travel demand forecasting and agent-based traffic simulation. This paper proposes a new sample-based method for synthetic population generation, which can be viewed as an alternative of the traditional Iterative Proportional Fitting. The method introduces bootstrapping techniques to compute a discrete copula function. Based on the copula function, associations among different attributes can be estimated and the population structure can be recovered. Experiments using actual Chinese national population data indicate that the new method can achieve the same level of accuracy as Iterative Proportional Fitting while acquire better results of the partial joint distributions.

Keywords: Advanced Driver Assistance Systems, Autonomous Driving, Autonomous Vehicles
Abstract: Automated vehicle (AV) technologies are rapidly maturing, and time line for their wider deployment is currently uncertain. Despite uncertainty these technologies are expected to bring about numerous societal benefits, such as enhanced traffic safety, improved mobility and reduced fuel emissions. In this paper, we propose a novel bottom-up approach to model various SAE levels on VISSIM in a two lane highway environment featuring an on-ramp. Our results indicate that mobility in SAE level 1 always exceeds that of SAE level 0, because the former has a consistently higher acceleration for given conditions. SAE level 2 provides more lateral stability and therefore less implied accidents than level 1 or 0 due to lower lateral deviations. For level 3, key consideration is to model the transition between human and system control. In SAE level 4 we model the operation of autonomous vehicles in Operational Design Domain (ODD) and transition to minimal risk conditions outside ODD. SAE level 5 overcomes impact of these transitions and hence has a better mobility than lower SAE levels. The models can help policymakers to understand the impact of autonomous vehicles on mobility and guide them in making critical policy decisions.

Keywords: Automated Driving, Safety, Hardware, software, and human-in-the-loop simulation
Abstract: The safety validation of automated driving for SAE level 3 and higher is still an unsolved issue. While it is generally assumed that an automated driving system should be at least as safe as a human driver, the proof of safety is currently not feasible applying approaches that are accepted for validation today. This paper gives an overview of existing approaches of safety validation from the stakeholders’ perspective. Subsequently, the approaches are structured systematically, thus highlighting the simplifications they suffer from. The advantages and drawbacks of each approach are summed up. It is concluded that the issue of safety validation cannot be “solved”, as in proven mathematically with absolute certainty. The goal of the international community needs to be to design the safety validation based on state-of-the-art approaches and then examine acceptance of these approaches. The acceptance needs to be examined from all stakeholders based on the identified pros, cons, and simplifications. Hence, several risk-handling strategies are introduced to deal with the uncertain safety level resulting from necessary simplifications.

Keywords: Autonomous Driving, Sensing, Vision, and Perception, Segmentation
Abstract: In this paper, a slope-robust cascaded ground segmentation in 3D point cloud for autonomous vehicles is presented. In many challenging terrains encountered by autonomous vehicles where the ground does not have a simple planar shape such as sloped roads, many existing ground segmentation algorithms fail. The proposed algorithm aims to correctly segment ground points in scans where these challenging terrains are present. The proposed method consists of two main steps. First, filtering the majority of non-ground points using the geometry of the sensor and the distance between consecutive rings in the scan. In the second step, multi-region RANSAC plane fitting is used to separate remaining non-ground points from ground points in the scan. The 3D data was taken and partially labeled for quantitative evaluation. The experimental results were outstanding as the proposed algorithm could segment the ground correctly in various challenging terrains. The proposed algorithm could correctly segment ground points in the scan even in sloped terrains and achieved higher accuracy than other algorithms used in the evaluation.

Keywords: Autonomous Vehicles, Automated Driving, Path Planning
Abstract: Automated driving systems are likely to encounter situations in which their understanding of the road is uncertain. State-of-the-art systems only act on a single road hypothesis. If that hypothesis is incorrect, the likely consequence is an undesired behavior of the automated vehicle. This paper shows that the consideration of multiple road hypotheses in the selection of a driving corridor can improve the performance of an automated driving system and relax the requirements on its perception system. Two algorithms are presented that infer a corridor from a multi-hypothesis road representation with respect to different navigational goals.

Keywords: .
Abstract:

Keywords: Routing Games, Simulation and Modelling, Traffic Congestion
Abstract: The rise of congestion across the United States and the increasing adoption of mobile routing services have enabled drivers with the ability to find the fastest routes available in urban road networks. Arterial roads and side streets originally designed for local traffic are impacted by the influx of selfishly routed drivers, garnering much recent media attention and civic debate. Classic flow-based game theoretic models provide the framework for simulating the behavior of routed and non-routed drivers on a road network. We developed an interactive policy decision support system called the Routing Impact Detection, Evaluation, and Response Decision Support System (RIDER DSS) as a tool for policymakers and practitioners to hone in on areas most impacted by routing apps and assess potential policy actions to mitigate the effects of cut-through traffic on a local and regional scale. In a case study of Baxter Street in the Los Angeles Basin we demonstrate how the RIDER DSS can relate the percentage of app users in a network to the distribution of traffic flow on side streets.

Keywords: Traffic Congestion, Traffic Flow Model, Urban Traffic Control
Abstract: This paper proposes a methodology to evaluate the performance of a road network during non-recurring congestion for real-time traffic control applications. A novel performance indicator based on the concept of the macroscopic fundamental diagram (MFD) is developed to assess the travel production of the network. The proposed indicator is obtained by calculating the weighted space-mean flow of an urban network, which is a proxy for the travel production of the corresponding network. The resilience of the road network is defined as its ability to retain the same level of travel production after occurrence of a disruption. This paper shows how real-time traffic control measures can enhance the resilience of the network. More specifically, the impact of re-routing as a real-time traffic management measure is investigated in a network where a link is closed due to an unpredictable incident. The main advantage of this approach in comparison to the existing travel time based approaches is that it neither requires a detailed model of the network nor a calibration of a dynamic traffic assignment model for different demand scenarios, as the MFD is a property of the network and is not sensitive to small changes in demand.

Keywords: Traffic Simulation, Congestion, Driver Modelling
Abstract: This paper analyzes the impact of late merging vehicles on traffic flow and energy efficiency via traffic microsimulation. Lane reductions, due to construction, road obstructions, or transitions to narrower roads, can be disruptive to traffic flow and energy efficiency of a stream of vehicles. In these instances, when there is heavy traffic flow a bottle-necking effect occurs at the reduction point. The backup of traffic is then propagated upstream causing vehicles to decelerate more often. As a result, fuel efficiency decreases and there is an increase in trip time. This research is motivated by recent articles that indicate merging late could be beneficial when done in a coordinated manner. This paper will focus on a group of vehicles traveling along a high speed roadway with a lane closure. These vehicles employ a reactive Intelligent Driver Model with randomized parameters that mimic human-like car following. Our microsimulations show that, in general, high penetrations of late merging vehicles have a negative influence on traffic flow and energy efficiency. We consider a typical lane change behavior that does not involve coordination which could be the reason our results contradict those found from the Late Merge method and other common lane change models.

Keywords: Travel Time, Traffic Congestion, Prediction
Abstract: We present a new method for long-term estimation of the expected travel time for links on highways and their variation with time. The approach is based on a time series analysis of travel time data from the UK's National Traffic Information Service (NTIS). Time series of travel times are characterised by a noisy background variation exhibiting the expected daily and weekly patterns punctuated by large spikes associated with congestion events. Some spikes are caused by peak hour congestion and some are caused by unforeseen events like accidents. Our algorithm uses thresholding to split the data into background and spike signals, each of which is analysed separately. The the background signal is extracted using spectral filtering. The periodic part of the spike signal is extracted using locally weighted regression (LWR). The final estimated travel time is obtained by recombining these two. We assess our method by cross-validating in several UK motorways. We use 8 weeks of training data and calculate the error of the resulting travel time estimates for a week of test data, repeating this process 4 times. We find that the error is significantly reduced compared to estimates obtained by simple segmentation of the data and compared to the estimates published by the NTIS system.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Trajectory Prediction, Connected Vehicles
Abstract: On arterial corridors, due to traffic signals and vehicle demands at all different approaches, traffic conditions can be very different across lanes. Vehicular queues and average speed at different lanes, especially lane groups with different turns, can be extremely different even under the same road segment. This paper develops an innovative vehicle routing system with the consideration of vehicle queues and delays as the turning costs to search for the optimal routes for individual vehicles. The algorithm applies connected vehicles to estimate turning costs and dynamically updates vehicle routes with the prediction of the costs at each intersection along the routes. The system is incorporated in the INTEGRATION microscopic traffic simulator to conduct a comprehensive evaluation. The results indicate that the proposed algorithm can reduce the average travel time of connected vehicles and entire networks by up to 49% at one congested grid network. The impact of market penetration rates of connected vehicles is also investigated in this paper.

Keywords: Autonomous Vehicles, Communications and Protocols in ITS, Connected Vehicles
Abstract: An innovative push for increased autonomy in vehicles has generated a variety of autonomous vehicle platforms and architectures. While motivated by the prospect of increased safety and efficiency of travel, there remains increased risks when autonomous agent operate and make decisions individually and in its own self-interest. The problem then becomes how to implement and scale autonomous vehicle technology, and structure the systems in such a way that it can keep pace with a rapidly changing world, benefiting not just individuals, but societies. This approach develops an Intelligent Transportation System (ITS) that relies on an infrastructure for sensing and control. The solution lies in the removal of sensing and high computational tasks from the vehicles, allowing static ground stations named Multi Sensor-Sensing Packs (MSSPs) containing the sensors, cameras, and computing to perceive the surrounding environment and navigate the vehicles safely. Testing a multi-nodal network of MSSPs determined wireless commands from the Infrastructure Enabled Autonomy (IEA) network can reliably operate a vehicle in a limited agent environment. Between the systems level vehicle control, network lag, and pixel error, the system remained robust enough to generate stable control and tracking of a vehicle on a planned path, avoiding obstacles with minimal error.

Keywords: Connected Vehicles, Automated Vehicles, Safety
Abstract: In this paper, we design a connected cruise con- troller with safety guarantees. In particular, we utilize a control safety function in order to guarantee the safety of a given control law. We establish the notion of safety chart to graphically represent the safe combinations of control parameters. Moreover, we establish an intervention algorithm that maintains safety when parameters are chosen outside the safe parameter regime. The results are also demonstrated with the help of numerical simulations.

Keywords: Connected Vehicles, Autonomous Driving, Simulation and Modelling
Abstract: This study aims to analyze the impact of connected and autonomous vehicles (CAVs) on traffic flow under various penetration rates. Based on a recently proposed heterogeneous flow model, the mixed traffic flow with both conventional vehicles and CAVs was simulated and studied. Acceleration rate and velocity distributions of the mixed traffic flow were presented to show the evolution of mixed traffic flow dynamics with the increase in CAV penetration rate within the mixed flow. Spatiotemporal diagrams of mixed traffic flow were presented to show the effect of CAVs on damping the stop-and-go traffic flow. Results show that with the increase in CAV penetration rate, the portion of smooth driving is increased. Velocity difference between vehicles is decreased and traffic flow is greatly smoothed. Stop-and-go traffic will be greatly eased when CAV penetration rate reaches a rate of 40%~50%. More cautious following strategy of the CAV would contribute to a greater benefit on traffic safety. This work provides some insights into the impact of the CAV on traffic flow and sheds light on how would the mixed traffic flow dynamics evolve with the gradual adoption of CAV under current traffic system.

Keywords: Sensing, Vision, and Perception
Abstract: Knowing is at least half the battle when it comes to building robust navigation techniques for real-world mobile robots. A robot’s knowledge is often derived from onboard sensors, but sometimes we forget that we as designers can choose any sensors we’d like to make things as easy as possible. This talk will provide several examples (from our lab over the last decade) of using some uncommon sensors/data to help robots understand the world including Sun/star trackers to help with localization, lidar intensity images (as opposed to pointclouds) and colour-constant images to deal with lighting change, temporal series of images to span drastic appearance change, and even a tether to determine where a robot is nonvisually. The talk will also comment on some other cool work going on beyond our lab and speculate a little about future promising directions for sensors.

Keywords: Perception, Computer Vision, Safety
Abstract: Moving obstacles occluded by corners are a potential source for collisions in mobile robotics applications such as autonomous vehicles. In this paper, we address the problem of anticipating such collisions by proposing a vision-based detection algorithm for obstacles which are outside of a vehicle's direct line of sight. Our method detects shadows of obstacles hidden around corners and automatically classifies these unseen obstacles as ``dynamic'' or ``static''. We evaluate our proposed detection algorithm on real-world corners and a large variety of simulated environments to assess generalizability in different challenging surface and lighting conditions. The mean classification accuracy on simulated data is around 80% and on real-world corners approximately 70%. Additionally, we integrate our detection system on a full-scale autonomous wheelchair and demonstrate its feasibility as an additional safety-mechanism through real-world experiments. We release our real-time capable implementation of the proposed ShadowCam algorithm and the dataset containing simulated and real-world data under an open-source license.

Keywords: Sensing, Vision, and Perception, Text Mining, Camera
Abstract: Traffic cameras are commonly deployed monitoring components in road infrastructure networks, providing operators visual information about conditions at critical points in the network. However, human observers are often limited in their ability to process simultaneous information sources. Recent advancements in computer vision, driven by deep learning methods, have enabled general object recognition, unlocking opportunities for camera-based sensing beyond the existing human observer paradigm. In this paper, we present a Natural Language Processing-inspired approach, entitled Bag-of-Label-Words (BoLW), for analyzing image data sets using exclusively textual labels. The BoLW model represents the data in a conventional matrix form, enabling data compression and decomposition techniques, while preserving semantic interpretability. We apply the Latent Dirichlet Allocation topic model to decompose the label data into a small number of semantic topics. To illustrate our approach, we use freeway camera images collected from the Boston area between December 2017–January 2018. We analyze the cameras’ sensitivity to weather events; identify temporal traffic patterns; and analyze the impact of infrequent events, such as the winter holidays and the “bomb cyclone” winter storm. This study demonstrates the flexibility of our approach, which allows us to analyze weather events and freeway traffic using only traffic camera image labels.

Keywords: Cooperative Automated Vehicles, Motion Planning, Trajectory Planning
Abstract: While motion planning approaches for automated driving often focus on safety and mathematical optimality with respect to technical parameters, they barely consider convenience, perceived safety for the passenger and comprehensibility for other traffic participants. For automated driving in mixed traffic, however, this is key to reach public acceptance. In this paper, we revise the problem statement of motion planning in mixed traffic: Instead of largely simplifying the motion planning problem to a convex optimization problem, we keep a more complex probabilistic multi agent model and strive for a near optimal solution. We assume cooperation of other traffic participants, yet being aware of violations of this assumption. This approach yields solutions that are provably safe in all situations, and convenient and comprehensible in situations that are also unambiguous for humans. Thus, it outperforms existing approaches in mixed traffic scenarios, as we show in our simulation environment.

Keywords: Cooperative Automated Vehicles, Traffic Signal Control, Collaboration, cooperation, competition, coalitions in traffic and transportation models
Abstract: This paper addresses the problem of optimally coordinating connected vehicles crossing an intersection without any explicit traffic signals. We propose a game-in-game framework that utilizes Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) technologies in order to maximize intersection throughput and to minimize traffic accidents and congestion. A Platoon Structure Formation Algorithm (PSFA) is proposed to form coalitions for CAVs at the intersection boundary, and a strategic game for CAVs in the interior is proposed to avoid predicted accidents inside the intersection. We perform extensive experiments to evaluate our proposed game-in-game framework under different traffic conditions. The results show that our proposed game-in-game framework reduces the accidents by 99%, while increasing the intersection throughput significantly.

Keywords: Trajectory Tracking, Bayesian Estimation, Road Traffic
Abstract: In many applications, tracking of multiple objects is crucial for the perception of the current environment. Most of the present multi-object tracking algorithms assume that objects move independently regarding to other dynamic objects as well as the static environment. Since in many traffic situations objects interact with each other and in addition there are restrictions due to drivable areas, the assumption of an independent object motion is not fulfilled. This paper proposes an approach adapting a multi-object tracking system to model interaction between objects and the current road geometry. Therefore, the prediction step of a Labeled Multi-Bernoulli filter is extended to facilitate modeling interaction between objects using the Intelligent Driver Model. Furthermore, to consider road map information, an approximated form of a highly precise road map is used. The results show that in scenarios where the assumption of a standard motion model is violated, the tracking system adapted with the proposed method achieves higher accuracy and robustness in its track estimations.

Keywords: Simulation and Modelling, Cooperative Driving, Connected Vehicles
Abstract: The field of cooperatively interacting vehicles requires complex simulation infrastructures dealing with various aspects such as vehicle, traffic, and communication models. In this work we present a modular and extensible simulator architecture, design patterns, and best practices for this domain. We show how extension points for co-simulators can be employed allowing the engineer to tailor a simulation environment to his needs. Moreover, we introduce the zoning approach distributing the computational burden of a simulation over a series of machines thereby allowing us to cope with a large number of participants residing in large urban areas.

Keywords: Cooperative Driving, Trajectory Planning, Motion Planning
Abstract: This paper considers the problem of cooperative trajectory planning for multiple, communicating, automated vehicles in general non-hazardous on-road scenarios. Cooperative behavior is introduced by optimizing a collective cost function for all automated vehicles. A novel approach based a Mixed-Integer Quadratic Programming is presented that is guaranteed to yield the globally optimal solution. Numerical experiments are provided to demonstrate the feasibility of our approach and the benefits compared to priority-based approaches and non-cooperative individual motion planning.

Keywords: Hardware, software, and human-in-the-loop simulation, Large scale simulation of agent-based traffic models, Simulation and Modelling
Abstract: Transit systems have encountered a radical change in the recent past as a result of the digital disruption. Consequently, traditional public transit systems no longer satisfy the diversified demands of passengers and hence, have been complemented by demand responsive transit solutions. However, we identify a lack of simulation tools developed to test and validate complex scenarios for real-time demand responsive public transit. Thus, in this paper, we propose a simulation framework, which combines complex scenario creation, optimization algorithm execution and result visualization using SUMO, an open source continuous simulator. In comparison to a state-of-the-art work, the proposed tool supports features such as varying vehicle capacity and driving range, immediate and advance passenger requests and maximum travel time constraints. Further, the framework follows a modular architecture that allows plug-and-play support for external modules.

Keywords: Traffic Flow Prediction, Data Mining and Data Analysis for Artificial Transportation Systems, Computational Intelligence
Abstract: Traffic prediction is an elemental function of Intelligent Transportation Systems, and accurate and timely prediction is of great significance to both traffic management agencies and individual drivers. With the development of deep learning and big data, deep neural networks (DNN) achieve superior performances in traffic prediction. Developing DNN prediction models needs large scale and diverse data, however, it is costly to collect large volume of accurate traffic data. In this paper, we propose to use small volume of real traffic data and large volume of synthetic traffic data to developing traffic prediction models. The evolving of parallel system paradigm for traffic prediction and the algorithm to incrementally train traffic data generation models and traffic prediction models are presented. We use an improved generative adversarial networks to generate traffic data, and a stacked long short-term memory model for traffic prediction. Experimental results on a real traffic dataset demonstrate that our method can significantly improve the performance of traffic flow prediction.

Keywords: Theory and Models for Optimization and Control, Traffic Signal Control, Traffic Simulation
Abstract: In order to reduce the adverse effects of the arterial intersections signal control strategy dynamic changes during arterial coordination control signal transition. This paper analyzes the fluctuation characteristics during the transition of control signals. The phase offset variation volatility model, signal cycle fluctuation volatility model, green ratio volatility model are constructed, and the volatility model analysis is conducted on the two aspects of the multi-period transition and multi-intersection coordination cycle transitions. The particle swarm optimization algorithm was used to solve the multi-objective optimization problem in the arterial coordination transition, and the best transition control strategy was finally obtained. The simulation results show that the proposed algorithm is optimized compared with the traditional Add algorithm and the Subtract algorithm in terms of vehicle average delay, vehicle average stop times, and average queue length, which can make the signal control effect better.

Keywords: Traffic Light Control, Traffic Congestion, Agent-based modelling and simulation
Abstract: In this paper we present DALI, a distributed, collaborative multi-agent Traffic Signal Timing system (TST) for highly dynamic traffic conditions. In DALI, intersection controllers are augmented with software agents which collaboratively adapt signal timings by considering the feedback of all controller agents that may be affected by a change. The model is based on a real-world TST and is intended to be deployed with minimal changes to the infrastructure. DALI has been validated on a simulated model of the City of Richardson, Texas, comprising 128 signalized intersections. The experimental results show that it outperforms the conventional traffic operation modes as well as an RL-based TST in highly dynamic scenarios.

Keywords: Large scale simulation of agent-based traffic models, Simulation and Modelling, Traffic Flow Model
Abstract: Transportation systems are becoming increasingly complex with the evolution of emerging technologies, including deeper connectivity and automation, which will require more advanced control mechanisms for efficient operation (in terms of energy, mobility, and productivity). Stakeholders, including government agencies, industry, and local populations, all have an interest in efficient outcomes, yet there are few tools for developing a holistic understanding of urban dynamics. Simulating large-scale, high-fidelity transportation systems can help, but remains a challenging task, due to the computational demand of processing massive numbers of events and the non-linear interactions between system components and traveling agents. In this paper, we introduce Mobiliti, a proof-of-concept, scalable transportation system simulator that implements parallel discrete event simulation on high-performance computers. We instantiated millions of nodes, links, and agents to simulate the movement of the population through the San Francisco Bay Area road network and provide estimates of the associated congestion, energy usage, and productivity loss. Our preliminary results show excellent scalability on multiple compute nodes for statically-routed agents, simulating 9.5 million trip legs over a road network with 1.1 million nodes and 2.2 million links, processing 2.4 billion events in less than 30 seconds using 1,024 cores on NERSC's Cori computer.

Keywords: Artificial Neural Network, Trajectory Tracking, Deep Learning
Abstract: This paper explores the capability of deep neural networks to capture key characteristics of vehicle dynamics, and their ability to perform coupled longitudinal and lateral control of a vehicle. To this extent, two different artificial neural networks are trained to compute vehicle controls corresponding to a reference trajectory, using a dataset based on high-fidelity simulations of vehicle dynamics. In this study, control inputs are chosen as the steering angle of the front wheels, and the applied torque on each wheel. The performance of both models, namely a Multi-Layer Perceptron (MLP) and a Convolutional Neural Network (CNN), is evaluated based on their ability to drive the vehicle on a challenging test track, shifting between long straight lines and tight curves. A comparison to conventional decoupled controllers on the same track is also provided.

Keywords: Automated Driving, Human-Machine Interface, Driver State
Abstract: At a level-3 or higher level automation [1], a driver does not have to constantly monitor the vehicle and environment while driving, which enables the driver to conduct non-driving-related tasks (NDRTs) and be out of the control loop. This may influence a driver’s visual behavior, cognitive states, which leads to loss of situation awareness (SA) and skills. This is dangerous if the automated system reaches its boundaries: the driver must take-over the driving task in a critical situation within a limited period of time. In this paper, a concise HMI concept of the LED ambient light positioned at the bottom of the windscreen is presented, which contains information about the status and intention of the automation, detected potential hazards and the warning for a take-over request (TOR) by varying the LED’s color, frequency, lighting position and animation. The goal is to increase the SA during automated driving and improve the take-over quality while allowing the driver to perform NDRTs without distraction and annoyance. In this between-subject-design experiment in a static driving simulator, 50 participants performed a visual-motoric task on a smartphone during a 45-min automated drive with or without the new HMI. Compared to the baseline, results show significant improvements in the gaze behavior and take-over quality. The new HMI also shows a high acceptance and increases the trust in automation while avoiding overtrust.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Deep Learning, Microscopic Traffic Simulation
Abstract: We consider long-term planning problems for autonomous vehicles in complex traffic scenarios where vehicles and pedestrians interact. The decisions of an autonomous vehicle can influence surrounding other participants in these scenarios. Therefore, planning algorithms that preprocess the long-term prediction of other participants restrict freedom in action. In this paper, we process both problems of long-term planning and prediction at the same time. Our approach which we call DDT (Deep Driving Tree) is based on game tree accumulating a short-term prediction. Machine learning techniques are applied to this short-term prediction instead of model-based techniques that depends on domain knowledge. In contrast to Q-learning, this prediction part is trained off-line and does not require feedback from collision data. Our approach using a game tree models multiple future states of other participants to decide a proactive action taking uncertainties of their intentions into consideration. This approach is demonstrated in a left turning scenario at an intersection of left-hand traffic with oncoming vehicles without V2V communication.

Keywords: Automated Vehicle Operation, Motion Planning, Navigation, Internet of Things, Traffic Management
Abstract: This paper presents an architecture for Automated Valet Parking (AVP) connected to cloud-based IoT services and mobile user interfaces. The goal is to enable AVP services for automatic vehicles. From the user perspective, automatic car drop-off and pick-up are activated via smart phone application, and the user will be able to continuously monitor the vehicle status together with additional services as cleaning or recharge during the parking phase. Further, the IoT platform allows the integration of live services that will interact with automatic driving and parking. As an example, the presented AVP setup includes the operation of service drones to automatically guide a vehicle to the best parking spot. The demonstration in this paper comprises a parking car and a micro aerial vehicle (MAV) connected in real-time through the IoT platform as well as the smart phone application where the car is controlled and supervised.

Keywords: Automated Vehicles
Abstract: Path planning is a crucial task in automated driving. Due to the complexity of dynamically changing driving environment, the planned path generally requires the capability to adjust itself in real time to avoid obstacles detected in its way. The use of an optimization method is able to generate a collision-free and smooth path. However, its high computation burden limits its direct application to online path planning. This paper proposed a time-efficient online table-lookup approach to deal with this dilemma. Given discrete target points, this approach is capable to form a quintic-spline path with second-order geometric (G2-) continuity using a look-up table. The look-up table was generated beforehand in the reference space with minimization on curvature variation. The paper demonstrates the application of this online approach in collision avoidance, with a geometry-based method to decide new target points when obstacles are detected in the original path. These new target points are fed to the table-lookup online path planning algorithm to generate a collision-free path with minimum curvature variation.

Keywords: .
Abstract:

Keywords: .
Abstract:

Keywords: .
Abstract:

Keywords: .
Abstract:

Keywords: .
Abstract:

Keywords: .
Abstract:

Keywords: .
Abstract: The speakers in this session, all former chairs or presidents of the ITS Society or its predecessor Committee and Council, will share their perspectives on the following topics: • Past. A brief history of the events and organizational structures leading up to the current ITS Society and their contributions and relationships with related organizations • Present. The current state of ITS applications and research globally • Future. The future of ITS applications and research and the ITS Society.

Panelists: • Christoph Stiller, IEEE ITS Society President (2012-13) • Umit Ozguner, IEEE ITS Council Chair (2000) • Chip White, IEEE ITS Committee Chair (1994-95)

Keywords: Destination Prediction, Bayesian Estimation, Driver Assistance Systems
Abstract: We present here a general framework and a specific algorithm for predicting the destination, route, or more generally a pattern, of an ongoing journey, building on the recent work of [Y. Lassoued, J. Monteil, Y. Gu, G. Russo, R. Shorten, and M. Mevissen, "Hidden Markov model for route and destination prediction," in IEEE International Conference on Intelligent Transportation Systems, 2017]. In the presented framework, known journey patterns are modelled as stochastic processes, emitting the road segments visited during the journey, and the ongoing journey is predicted by updating the posterior probability of each journey pattern given the road segments visited so far. In this contribution, we use Markov chains as models for the journey patterns, and consider the prediction as final, once one of the posterior probabilities crosses a predefined threshold. Despite the simplicity of both, examples run on a synthetic dataset demonstrate high accuracy of the made predictions.

Keywords: Deep Learning, Advanced Driver Assistance Systems, Classification
Abstract: With the steady rise of advanced driver assistance systems (ADAS), more and more aspects of the driving task are transferred from the human driver to the vehicle’s control system. In order to handle many of these responsibilities, vehicles need to understand their environment and adjust their behavior according to it. An important aspect of the vehicle environment is the layout of the road segment right ahead of the vehicle, such as the presence and type of an intersection, as it defines the scenario, provides context information and constrains the future motion of traffic participants. The knowledge of upcoming intersections can help to improve various aspects in the context of driver assistance systems and automated driving, such as the prediction of traffic participants or the adjustment of a system with respect to the current scenario. The contribution of this paper is threefold: First, it introduces a model for intersection identification and classification ahead of a vehicle solely from on-board sensor data via deep learning. Second, it proposes a transfer-learning technique allowing to train with fewer samples and showing that intermediate features from path prediction are also beneficial for intersection classification tasks. Third, it allows to reduce necessary computational power since feature extraction is partially shared between the path prediction and the intersection classification model.

Keywords: Driver Modelling, Advanced Driver Assistance Systems, Artificial Neural Network
Abstract: Prediction of human driving decisions is an important aspect of modeling human behavior for the application to Advanced Driver Assistance Systems (ADAS) in the intelligent vehicles. This paper presents a sensor based receding horizon model for the prediction of human driving commands. Human driving decisions are expressed in terms of the vehicle speed and steering wheel angle profiles. Environmental state and human intention are the two major factors influencing the human driving decisions. The environment around the vehicle is perceived using LIDAR sensor. Feature extractor computes the occupancy grid map from the sensor data which is filtered and processed to provide precise and relevant information to the feed-forward neural network. Human intentions can be identified from the past driving decisions and represented in the form of time series data for the neural network. Supervised machine learning is used to train the neural network. Data collection and model validation is performed in the driving simulator using the SCANeR studio software. Simulation results are presented along with the analysis.

Keywords: Driver Monitoring, Advanced Driver Assistance Systems, Convolutional Neural Networks
Abstract: The ability to monitor the visual attention of a driver is a useful feature for smart vehicles to understand the driver's intents and behaviors. The gaze angle of the driver is not deterministically related to the head pose of the driver due to the interplay between head and eye movements. Therefore, this study aims to establish a probabilistic relationship using deep learning. While probabilistic regression techniques such as Gaussian process regression (GPR) has been previously used to predict the visual attention of a driver, the proposed deep learning framework is a more generic approach that does not make assumptions, learning the relationship between gaze and head pose from the data. In our formulation, the continuous gaze angles are converted into intervals and the grid of angles is treated as an image for dense prediction. We rely on convolutional neural networks (CNNs) with upsampling to map the six degrees of freedom of the orientation and position of the head into gaze angles. We train and evaluate the proposed network with data collected from drivers who were asked to look at predetermined locations inside a car during naturalistic driving recordings. The proposed model obtain very promising results, with 11.73% surface area of a sphere containing 95% of the true gaze angles in the test set. The architecture offers an appealing and general solution to convert regression problems into dense classification problems.

Keywords: Sensing, Vision, and Perception, Theory and Models for Optimization and Control, Advanced Driver Assistance Systems
Abstract: Automation of complex traffic scenarios is expected to rely on input from a roadside infrastructure to complement the vehicles' environment perception. We here explore design requirements for a prototypical setup of virtual vision or RADAR sensors along one roadside. Explicitly, we analyze the road coverage and the probability of vehicle occlusions, with the objective of evaluating the completeness of information that is captured by the sensor field. Simulation case studies are performed based on real traffic data acquired at the German Autobahn 9 near Munich. Our findings indicate how the sensor network should be designed in terms of sensor range, orientation and opening angle, in order to enable effective traffic detection. The achieved degree of completeness suggests that such a setup could be used to support automated vehicles to a substantial extent.

Keywords: Advanced Driver Assistance Systems, Clustering, Classification
Abstract: Traffic light recognition is of great interest for advanced driver assistance systems and autonomous driving but still an unsolved problem. While a traffic light has few visual features for detection from camera images we believe the characteristic light represents a potentially very strong and stable feature. The traffic light is actively emitting light which is rarely influenced by weather or lighting condition. When using a color lookup table for an image segmentation-based object detector, the process of creating the lookup table is the crucial point. In this paper, we propose a method for generating a lookup table using real world data of a large dataset. The training data is sampled from labeled objects and stored as multisets. We contribute a frequency-based filtering method to clean the samples before using a k-nearest neighbor classifier to generalize. The result is stored as a three dimensional lookup table. The main contribution is a neighborhood-biasing technique that allows setting different operating points online without retraining. A challenging real world dataset containing several thousands of active lights is used to evaluate the process.

Keywords: Human Factors, Driving Style, Driver Behaviour
Abstract: This paper presents an approach for predictive driving when facing blind intersections based on expert data. Expert drivers anticipate and avoid potential dangerous situations. In most cases these complex behaviors cannot be reproduced by state of the art planning approaches. We present an analysis of expert drivers while passing through blind intersections, we extract useful driving features to model the intersection and propose a cost function based on those features. We use inverse reinforcement learning to extract a feature-based cost function and learn its parameters from driver data. Finally, feature weights are computed based on collected expert driving data (using 211 trajectories). Evaluation was performed in terms of trajectory and speed using modified Hausdorff distance. Experimental results show that the planner is capable of computing trajectories comparable to those ones of the expert drivers when facing blind intersections.

Keywords: Speed Trajectory Optimization, Cooperative Adaptive Cruise Control, Human Factors
Abstract: Autonomous vehicles are expected to start reaching the market within the next years. However in practical applications, navigation inside dynamic environments has to take many factors such as speed control, safety and comfort into consideration, which is more paramount for both passengers and pedestrians. In this paper, a novel speed profile planner based on an optimal control approach considering passenger comfort is proposed. The approach is accomplished by minimizing jerk under certain comfort constraints, which inherently gives a speed profile for the central nervous system to follow naturally. Imposed with the same conditions, the widely used Jerk Limitation method is interpreted as an equivalent of the minimum time control method, the latter being used to verify that our method can ensure better continuity and smoothness of the speed profiles. A validation test was specifically designed and performed in order to show the feasibility of our method.

Keywords: Agent-human interactions, Social transportation, Autonomous Driving
Abstract: Cooperative interaction in traffic is vital for resolving a wide range of ambiguities arising from road users' actions. Autonomous vehicles are no exception and require the ability to understand the intention of road users and communicate with them in order to ensure their safety and maintain traffic flow.

In this paper, we address the problem of traffic interaction by analyzing a large sample of pedestrians communicating with drivers. We highlight the ways pedestrians communicate and use a logistic regression model to identify what factors influence communication patterns of pedestrians and how. We also discuss practical challenges regarding the recognizing and understanding of pedestrians' intention and how our theoretical findings can help to solve them.

Our analysis suggests that pedestrians predominantly rely on implicit communication cues such as stepping onto the road to transmit their intention of crossing. In addition, we found that the presence of traffic signal, street width, and pedestrian group size can influence the frequency and type of pedestrian communication, while factors such as pedestrians' age and gender did not show any significant impact.

Keywords: Driver Modelling, Road Safety, Automated Driving
Abstract: Concerning the market introduction of automated driving, one of the major challenges for many years to come is to identify risks and benefits of these functions. Since automated driving functions are addressing a huge amount of relevant driving situations by means of their complex decision making algorithms, their effectiveness has to be determined in all possible driving situations in their operational design domain. Besides, the automated driving functions need to be assessed with respect to human driving performance according to the guidelines defined by the German Ethics Commission for Automated and Connected Driving [1]. Thus, based on a methodology for safety performance assessment of automated driving functions, this paper introduces specific driver models for modelling of human driver performance. How these models have to be structured and parameterized in order to present the variety of human driver performance under different situational variations is investigated in this paper. For validation of the derived models, a study in a highly dynamic driving simulator on the example of a near-crash cut-in driving situation was carried out.

Keywords: Learning and adaptation, Deep Learning, Driving Style
Abstract: Modeling the driver’s behavior is an importance task which has shown its advantages in improving the prediction accuracy of different applications in the automotive field. A model of driver behavior could be used to increase safety, to improve the functionality of an advanced driver assistant system (ADAS), as well as to increase the driving experience and comfort for users. Most data-driven approaches build a general model that works well for the majority of drivers in the training set. Personalization, on the other hand, addresses the problem of adapting the model on the driver. A personalized model also has to be able to adjust to changing user preferences over time. In this work, we formulate the problem of personalization in automotive scenes, and propose an approach to extract and incorporate information from previous maneuver execution to improve the performance of prediction tasks in impending maneuvers. We apply our proposed adaptation method to predict the gap taken at a left-turn scenario and show that it can boost the prediction of a neural network in terms of F1 score in comparison to the baseline method by about 7% using fully connected layer, and by more than 9% when using LSTM layer.

Keywords: Decision Modeling, Bayesian Estimation
Abstract: In this paper, we implement an information-theoretic approach to travel behaviour analysis by introducing a generative modelling framework to identify informative latent characteristics in travel decision making. It involves developing a joint tri-partite Bayesian graphical network model using a Restricted Boltzmann Machine (RBM) generative modelling framework. We apply this framework on a mode choice survey data to identify abstract latent variables and compare the performance with a traditional latent variable model with specific latent preferences -- safety, comfort, and environmental. Data collected from a joint stated and revealed preference mode choice survey in Quebec, Canada were used to calibrate the RBM model. Results show that a significant impact on model likelihood statistics and suggests that machine learning tools are highly suitable for modelling complex networks of conditional independent behaviour interactions.

Keywords: Congestion, Network Modelling, Travel Time
Abstract: In order to stimulate the holiday economy and ensure smooth travel, the Chinese government decided to exempt the tolls of small passenger cars on major holidays since the National Day on October 1st, 2012. Taking a toll station in Shanghai as a case study, this paper analyzed its queuing situation on the first day of Tomb-sweeping day holiday before and after the toll-free policy with a queuing theory model. How to balance the induced traffic and the limited highway capacity with optimal toll strategies was also discussed. The results show that with total departure traffic volume increasing and an average service time decreasing, the average queue length and congestion duration became longer for the first year after the implementation of the policy. But then queuing situations were improved markedly in later years with increasing percentage of ETC users. The conclusions may benefit policy makers for transition process of different toll strategies around the world.

Keywords: Deep Reinforcement Learning, Cooperative Adaptive Cruise Control, Congestion
Abstract: Using deep reinforcement learning, we derive novel control policies for autonomous vehicles to improve the throughput of a bottleneck modeled after the San Francisco-Oakland Bay Bridge. Using Flow, a new library for applying deep reinforcement learning to traffic micro-simulators, we consider the problem of improving the throughput of a traffic benchmark: a two-stage bottleneck where four lanes reduce to two and then reduce to one. We first characterize the inflow-outflow curve of this bottleneck without any control. We introduce an inflow of autonomous vehicles with the intent of improving the congestion through Lagrangian control. To handle the varying number of autonomous vehicles in the system we derive a per-lane variable speed limits parametrization of the controller. We demonstrate that a 10% penetration rate of controlled autonomous vehicles can improve the throughput of the bottleneck by 200 vehicles per hour: a 25% improvement at high inflows. Finally, we compare the performance of our control policies to feedback ramp metering and show that the AV controller provides comparable performance to ramp metering without the need to build new ramp metering infrastructure. Illustrative videos of the results can be found at https://sites.google.com/view/itsc-lagrangian-avs/home and code and tutorials can be found at https://github.com/flow-project/flow.

Keywords: Free Way Traffic Control, Traffic Congestion, Simulation and Modelling
Abstract: The advent of automated, infrastructure-controlled vehicles offers new opportunities for traffic control. Even when the number of controlled vehicles is small, they can significantly affect the surrounding traffic. One way of regulating traffic is by using the automated vehicles as controlled moving bottlenecks. We present an extension of the cell transmission model that includes the influence of moving bottlenecks, consistently with the corresponding PDE traffic model. Based on this model, a control strategy is derived for traffic jam resolution. The strategy is tested in simulations, and shown to reduce the average travel time of surrounding vehicles, while also helping dissipate the traffic jam faster and ensuring the controlled vehicle avoids it.

Keywords: Road Safety, Traffic Congestion, Traffic Management
Abstract: Many methodologies have been developed to rank entities. The goal of developing ranking methodologies is to provide a means for decision-making. The current issue with ranking is that most results in terms of ranks are based on only one method. To resolve this issue, this paper proposes a hybrid model approach in the area of transportation and traffic safety. The hybrid model consists of the Normalized Score Summation Method, Principal Component Analysis, and the similarity measure based on the Proportion Discordance Ratio. The goal of this paper is to ensure the consistency of results from the three methodologies mentioned above. Its numeric analysis shows that results based on the hybrid model are more realistic and robust than the ones solely based on only one method.

Keywords: Path Planning, Traffic Congestion, Shared Mobility Systems
Abstract: Traffic congestion is a huge problem for transportation service providers, which causes delay and incurs excessive operational cost. Many algorithms exist for the dial-a-ride problem (DARP), which assumes constant travel times throughout the day. This may cause serious constraint violations when they are applied in the real-world, where time-varying travel times are observed. This paper aims to solve time-dependent dial-a-ride problem (TDDARP). A new greedy ant colony optimization (GACO) algorithm is developed, in which two new decision factors (repeat counter and quantity counter) are introduced to efficiently explore the search space. In addition, a method for estimating travel time using staircase regression (SR) of speed data is developed, which in turn is incorporated into GACO. Computational experiments conducted on several DARP benchmark instances in the literature show the significance of incorporating time-dependent travel times into designing high quality solution for TDDARP, while considering peak hour traffic congestion. On average, GACO attains solutions with 88.77% less time window constraint violation and 98.04% less ride time constraint violation after taking time-dependent travel times into account.

Keywords: Communications and Protocols in ITS
Abstract: Currently, the measures to traffic accident are advanced with the spread of the automobile. This paper discusses the pedestrian positioning system based on UWB(Ultra-Wide Band) ranging. In this system, base stations to receive UWB signal that is transmitted from pedestrians are attached to a traffic light for pedestrians. Positions of pedestrians are estimated by LSM(Least Square Method) using the ranging value that is estimated by UWB ranging scheme. The slotted ALOHA scheme is adopted for multiple access to access control. UWB positioning estimation system can detect positions with the error of 40cm. However, detailed characteristics of UWB positioning estimation system is not analyzed. In this paper, positioning error and PSR(Positioning Successful Rate) of estimation system are evaluated and analyzed by computer simulations. Also, positioning error and PSR are depended on the number of base stations. Thus, the effect of increasing the number of base stations is evaluated and analyzed. As the results of computer simulations, it is shown that the position of the pedestrian can be accurately estimated by using UWB positioning system.

Keywords: Communications and Protocols in ITS, V2v Communication, Autonomous Driving
Abstract: Despite years of development and deployment, the standardized IEEE802.11p communication for vehicular networks can be pushed toward insatiable performance demands for wireless network data access, with a remarkable increase of both latency and channel congestion levels when subjected to scenarios with a very high vehicle density. In some hard safety applications such as convoys, IEEE802.11p could seriously fail to meet the fundamental vehicular safety requirements. On the other hand, the advent of LED technologies has opened up the possibility of leveraging the more robust Visible Light Communication (VLC) technology to assist IEEE802.11p and provide seamless connectivity in dense vehicular scenarios. In this paper, we propose and validate a prediction-based vertical handover (PVHO) between VLC and IEEE802.11p meant to afford seamless switching and ensure the autonomous driving safety requirements. Algorithm validation and platoon system performance were evaluated using a specially implemented 802.11p-VLC module in the NS3 Network Simulator. The simulation results showed a speed-based dynamic redundancy before and after VLC interruptions with seamless switching. Moreover, the deployment of VLC for platoon intra-communication can achieve a 10-25% PDR gain in high-density vehicular scenarios.

Keywords: Connected Vehicles, Communications and Protocols in ITS, Connected Shared Mobility
Abstract: Mobility service platforms (MSPs) offer information, routing and booking of intermodal and personalized itineraries. Besides traditional travel information, MSPs can also provide sophisticated travel assistance actively supporting users during their itinerary and rerouting them in case of service disruptions. MSPs are usually accessed through the travelers smartphone, which unfortunately cannot be operated while driving due to safety regulations. As a consequence, users are disconnected while using their private or a rented/shared vehicle. In this work, we describe how to enable travelers' access to MSPs using the vehicle information system to allow a seamless travel assistance. In addition, the integration of the vehicle into the MSP also allows additional services, e.g., a personalized driving experience by synchronizing the users preferences between different vehicles. Our approach extends existing system architectures and communication protocols based on use cases for an intermodal journey. We prove the technical feasibility of our concept by demonstrating a working prototype implementation.

Keywords: Information Security and Privacy, Cryptographic Algorithms and Protocols, Communications and Protocols in ITS
Abstract: ETCS (European Train Control System) is an automatic control system that controls the speed limits of a train. Correct operation of this system is crucial as malfunctions in the system can cause serious accidents such as train collisions. It is essential that train-ground communication is reliable as this ensures the smooth operation of trains and accurate train parking on stations. The ground-based balise is a passive device that is energized by a passing train, and then communicates with the BTM (Balise transmission module) attached on the train via telegram messages to update the train of its location. However, there are vulnerabilities present in the balise air-gap interface that can be exploited by malicious attackers to alter or manipulated information on telegrams. This paper describes malicious attacks such as changing the locality information of balise can be exploited from attacking the vulnerability of air-gap interface of the balise-based system. This paper therefore proposes three security designs to check data integrity on telegram messages so as to protect the train balise telegram data integrity.

Keywords: V2v Communication, Communications and Protocols in ITS
Abstract: Due to high costs at initiation, alternative communication strategies to dedicated short range communication (DSRC) have been looked at to facilitate quick deployment of intelligent transportation services. Wi-Fi Direct (WD), has come across as a candidate, and this paper focuses on the drawback of large transmission delays in WD. To this end, it studies the performance gains of using a broadcast mechanism on the downlink between the group owner (GO) and the clients of a WD group, instead of the currently used peer-to-peer method. Gains in terms of average total delay and the average energy consumption of the GO are presented using a theoretical analysis as well as a simulation study using OMNeT++. It is also shown that the degradation in performance on the downlink due to not having retransmissions is within tolerable limits given that the size of the group is selected properly.

Keywords: V2v Communication, Connected Vehicles, Car-Following Model
Abstract: In this paper we propose a framework for experimentally evaluating the dynamics of connected vehicle networks using production vehicles. Connected vehicle networks contain human-driven connected vehicles which use wireless vehicle-to-vehicle (V2V) communication to transmit and receive messages, and connected automated vehicles that may use the received information to control their longitudinal motion. We use the developed framework to perform experiments on a connected vehicle network featuring two connected vehicles and one connected automated vehicle on public roads. The experiments show that by letting the connected automated vehicle utilize long-range connections we can benefit the traffic flow in connected vehicle networks. Lastly, we demonstrate that an analytical model of the connected vehicle network with matched parameters is able to reproduce the experimental results.

Keywords: Transport Health Management, Recurrent Neural Network, Prediction
Abstract: Road pavements need to be efficiently maintained under budget constraints. A pavement management system supports a road administrator’s decision making based on the prediction of pavement deterioration. However, the prediction of pavement deterioration is complicated and uncertain because there are many unobservable variables, and the highly accurate prediction of deterioration is difficult. For pavement administrators to use such predictions in decision making, it is necessary to quantify the reliability of prediction. This paper proposes a prediction-interval estimation method by applying the bootstrap method with a reduced computational cost to the degradation prediction model using a neural network. The proposed method is applied to the rutting depth in the inspection history of road pavement surfaces (i.e., inspection data of pavement surfaces), and the estimation accuracy of the prediction interval is verified. Verification shows that not only is the computational cost reduced but also the accuracy of the prediction interval is higher than that of the conventional method.

Keywords: Transport Health Management, Sensor Fusion, Connected Vehicles
Abstract: High-resolution road Situational Awareness (SA) will soon become a critical requirement for multitude of applications in next-gen intelligent transportation systems where a large number of autonomous vehicles are expected to be present. For instance, ride comfort, which is adversely affected by road anomalies such as potholes, is a much more pronounced concern for autonomous vehicles (where humans are driven) compared to traditional automobiles (being driven by humans). Towards developing efficient and cost-effective solutions for generating timely road condition data, this paper describes a target-tracking formulation of road anomaly detection by exploiting ubiquitous sensing and processing capabilities in modern automobiles. In particular, the focus is on large-scale monitoring of road networks by utilizing voluntarily participating motor-vehicles to detect and track the type and evolution of numerous road anomalies. The approach is described via an illustrative example on pothole detection and monitoring, where vibration and GPS data recorded by voluntarily participating motor-vehicles operating as ubiquitous sensors were utilized for detection and tracking.

Keywords: Travel Behaviour Under ITS, Road Traffic Management, Data Mining and Data Analysis
Abstract: The variable message signs, abbreviated as VMS messages, are disseminated through LED displays to provide the travelers and motorists information, warning and guidance on the current traffic situation. As an advanced traffic guidance system, the VMS messages help drivers to choose the routes with lower traffic volumes. Thus, the vehicles can be more evenly distributed in the road network to improve the performance of traffic system and reduce traffic delays. To this end, the VMS technology has been widely used in the expressways of Singapore. This paper aims to evaluate the immediate impact of VMS on the overall traffic distribution of Singapore in response to accidents and obstacles. For this purpose, we consider the incidents data and their corresponding VMS messages from the two busiest expressways of Singapore, namely Pan Island Expressway (PIE) and Central Expressway (CTE). For this analysis, we ignore the VMS messages of the locations which are already congested due to traffic incidents, since the motorists may be influenced by the congestion and not entirely by the VMS messages. The next step is to obtain the locations of other VMS displays and their nearby downstream exit points. The central argument of this analysis is that if the average traffic flow of the exits increases significantly compared to that of normal days, it demonstrates the impact of the VMS messages on the drivers’ behavior. Our results show that the average outgoing traffic flow from the expressways towards the exits increases by 14% after the VMS messages have been activated.

Keywords: Autonomous Vehicles, Detection, Classification
Abstract: We implement a speed-limit-sign recognition task using a template-based approach on the FPGA using the Intel FPGA SDK for OpenCL. Then we evaluate its throughput, power consumption, accuracy, and development effort against a GPU implementation that is based on a system presented in our previous study. This paper also discusses implementation differences between the FPGA and GPU systems, provides a methodology for translating the GPU system to the FPGA system, and explains optimizations used in the FPGA version. While implementing the FPGA implementation, we build an efficient FFT engine for image processing on the FPGA which can be utilized by other developers to perform related tasks. In this paper, we also provide our insights on building the FPGA versus GPU system, which we hope can be useful for designing upcoming versions of FPGA-focused OpenCL development environments. We conclude that the FPGA implementation provides better power consumption for the same detection accuracy, while the GPU supports better programmer efficiency.

Keywords: Automated Driving, Detection, Sensing, Vision, and Perception
Abstract: Robust traffic light detection and state recognition is of crucial importance on the path to automated vehicles. However, the mere classification of the signaled states does not suffice at complex multi-lane intersections. Rather, a complete understanding of the intersection, but at least the recognition of additional information (like arrows displayed on the traffic lights) is necessary. In this work, we developed a unified deep convolutional traffic light recognition system on the basis of the Faster R-CNN architecture, which is able to not only detect traffic lights and classify their state, but also distinguish their type (circle, straight, left, and right). An in-depth analysis of its performance on the large and diverse DriveU Traffic Light Dataset shows an overall detection performance of 0.92 Average Precision for traffic lights of width greater than 8 px. Additionally, other kinds of traffic lights, e.g. pedestrian lights, have been identified as main cause of false positives. Moreover, we evaluated the usefulness of the developed system to assess the traffic light states for all present driving directions revealing inconsistencies among multiple detections in single images.

Keywords: Multipath, Adas, Perception
Abstract: Future use of high-resolution near range radar sensors for vehicle environment perception is facing challenges in terms of detection and correct assignment of multipath reflections (false-positives) from surfaces and obstacles. This paper presents a novel geometric model to determine the relative positions from surrounding targets and reflection surfaces assuming that every object moves on a circular path to a mutual center. Principles of electromagnetic wave propagation under consideration of incident wave angles at potential reflection surfaces are described. Radar measurements illuminating an experimental target and a highway barrier as reflection surface, which reproducibly generates several multipath reflections, were carried out in a deterministic test environment and validate our presented methods. The installation of absorption material at intense reflection areas was conducted as countermeasure to reduce the intensity of false-positive detections. Furthermore, a realistic urban driving scenario using a real vehicle as target object and a building wall was reconstructed to proof field relevance.

Keywords: Intelligent Logistics, Path Planning, Artificial Intelligence
Abstract: This paper focuses on modeling and solving a last-mile package delivery routing problem with third-party drop-off points. The study is applicable to small or medium-sized delivery companies, which use bikes for performing the routes in an influence area bounded to a city. This routing setup has been formulated as a multi-objective optimization problem, balancing three conflicting objectives: a weighted measure of distance of the route, the safety of the biker, and the economic profit yielded by the delivery of goods to customers. Six different and heterogeneous multi-objective algorithms have been applied to the modeled problem: NSGA-II, MOCell, SMPSO, MOEA/D, NSGA-III and MOMBI2. In order to evaluate the performance of these algorithms, we have devised three experimental setups encompassing different real localizations in Madrid (Spain). For deploying a realistic simulation platform, the open-source Open Trip Planner framework has been used as a proxy evaluator of the produced routes. Results have been compared using the obtained Median and Inter Quartile Range of the hypervolume values reached by the algorithms. Conclusions drawn from this study show that MOCell is the best method for the proposed problem, reaching routes that balance the considered three objectives in a more Pareto-optimal fashion than the rest of counterparts in the benchmark.

Keywords: Public Transportation Management, Clustering
Abstract: Service regularity is one of the most significant performance indicators for public transport routes, typically measured through headway adherence. When headway deviations become too large and corresponding headways very small, bus bunching typically occurs. In these cases, passengers experience larger waiting times and overcrowding and an overall poor level of service. This paper aims to gain insight on frequent patterns of bus bunching using spatial analytics. Local and global spatial autocorrelation tests are performed on real world Automatic Vehicle Location (AVL) data to investigate spatial structures in the data. The spatio-temporal variations of bus bunching patterns throughout the day are further modeled using the ST-DBSCAN algorithm. Results show that the last few stops of each route exhibit statistically significant spatial autocorrelation with respect to the frequency of bunching, while the duration of bunching events is longer for route segments located in the central business district. Spatio-temporal clustering indicates that bunching is observed at a higher number of stops during peak traffic periods.

Keywords: Traffic Signal Control, Deep Reinforcement Learning, Reinforcement Learning
Abstract: Traffic signal control can mitigate traffic congestion and reduce travel time. A model-free reinforcement learning (RL) approach is a powerful framework for learning a responsive traffic control policy for short-term traffic demand changes without prior environmental knowledge. Previous RL approaches could handle high-dimensional feature space using a standard neural network, e.g., a convolutional neural network; however, to control traffic on a road network with multiple intersections, the geometric features between roads had to be created manually. Rather than using manually crafted geometric features, we developed an RL-based traffic signal control method that employs a graph convolutional neural network (GCNN). GCNNs can automatically extract features considering the traffic features between distant roads by stacking multiple neural network layers. We numerically evaluated the proposed method in a six-intersection environment. The results demonstrate that the proposed method can find comparable policies twice as fast as the conventional RL method with a neural network and can adapt to more extensive traffic demand changes..

Keywords: Deep Learning, Convolutional Neural Networks, Detection
Abstract: Window-sliding/region-proposal based methods have been the popular approaches for object detection with deep convolutional neural networks. However, these methods are very inefficient when the input image size is large, such as pavement images (2000times4000-pixel) used for cracking detection. In this paper, we propose a solution to this problem by introducing a fully convolutional dense-dilation network and the corresponding training strategy. The network is trained with small image blocks, then works on full-size images, which only needs to forward once for the process. In the first phase, it trains a classification network which classifies a small image block as crack, sealed crack or background. In the second phase, the fully convolutional layer is employed to convert the classification network into a detection network that is insensitive to the input size. At last, via introducing the equivalent dense-dilation design, it transfers both the low-level and middle-level knowledge from the classification network to facilitate the end-to-end network refining and improve the crack localization accuracy. The proposed approach is validated on 600 pavement images (2000times4000-pixel) obtained by industry equipment and it achieves state-of-the-art performance comparing with that of the recently published works in efficiency and accuracy.

Keywords: Localization, Camera, Sensing, Vision, and Perception
Abstract: For accurate vehicle self-localization, many approaches rely on the match between sophisticated 3D map data and sensor information obtained from laser scanners or camera images. However, when depending on highly accurate map data, every small change in the environment has to be detected and the corresponding map section needs to be updated. As an alternative, we propose an approach which is able to provide map-relative lane-level localization without the restraint of requiring extensive sensor equipment, neither for generating the maps, nor for aligning map to sensor data. It uses freely available crowdsourced map data which is enhanced and stored in a graph-based relational local dynamic map (R-LDM).

Based on rough position estimation, provided by Global Navigation Satellite Systems (GNSS) such as GPS or Galileo, we align visual information with map data that is dynamically queried from the R-LDM. This is done by comparing virtual 3D views (so-called candidates), created from projected map data, with lane geometry data, extracted from the image of a front facing camera. More specifically, we extract explicit lane marking information from the real-world view using a lane-detection algorithm that fits lane markings to a curvilinear model. The position correction relative to the initial guess is determined by best match search of the virtual view that fits best the processed real-world view.

Evaluations performed on data recorded in The Netherlands show that our algorithm presents a promising approach to allow lane-level localization using state-of-the-art equipment and freely available map data.

Keywords: Classification, Sensing and Intervening, Detectors and Actuators, Traffic Management
Abstract: Upcoming Intelligent Transportation Systems (ITSs) will transform roads from static resources to dynamic Cyber Physical Systems (CPSs) in order to satisfy the requirements of future vehicular traffic in smart city environments. Upto- date information serves as the basis for changing street directions as well as guiding individual vehicles to a fitting parking slot. In this context, not only abstract indicators like traffic flow and density are required, but also data about mobility parameters and class information of individual vehicles. Consequently, accurate and reliable systems that are capable of providing these kinds of information in real-time are highly demanded. In this paper, we present a system for the classification of vehicles based on their radio-fingerprints which applies cutting-edge machine learning models and can be non-intrusively installed into the existing road infrastructure in an ad-hoc manner. In contrast to other approaches, it is able to provide accurate classification results without causing privacy-violations or being vulnerable to challenging weather conditions. Moreover, it is a promising candidate for large-scale city deployments due to its cost-efficient installation and maintenance properties. The proposed system is evaluated in a comprehensive field evaluation campaign within an experimental live deployment on a German highway, where it is able to achieve a binary classification success ratio of more than 99% and an overall accuracy of 89.15% for a fine-grained classification task with nine different classes.

Keywords: Detection, Sensing and Intervening, Detectors and Actuators, Traffic Management
Abstract: In the ITS (intelligent transportation system), vehicle detection is one of the core technologies. We are developing an acoustic vehicle detector that detects vehicles using a sound map, which is a map of sound arrival time difference on two microphones. We developed vehicle detection algorithms based on state machine and DTW (dynamic time warping) to detect S-curves on a sound map drawn by passing vehicles. However, the detection algorithms often fail to detect simultaneous and sequential passing vehicles.

This paper presents SAVeD, a sequential acoustic vehicle detector. The SAVeD fits an S-curve model to sound map points using a RANSAC (random sample consensus) robust estimation method to detect each vehicle. The SAVeD then removes sound map points corresponding to the detected vehicle and continues vehicle detection process for the following vehicles. Experimental evaluations demonstrated that the SAVeD improves detection accuracy by more than 10 points compared to the state-machine based algorithm.

Keywords: Driver Behaviour, Driver Modelling, Intelligent Vehicles
Abstract: Driver identification constitutes an important enabling technology in intelligent transportation systems, allowing the development and the use of in-car personalised functionalities and thwarting unauthorised usage. In this work, we leverage the literature in authentication tasks (e.g. speaker recognition) and present a framework for driver identification which employs Support Vector Machine (SVM) and Universal Background Model schemes. Our framework operates on accelerator and break pedal signals, and thus augments other technologies, such as microphones or cameras, if present. Moreover, our framework is compatible with vehicles which are limited to traditional sensing modalities. We evaluate the framework on 15 hours of driving data for a total of 416 Km travelled, comprising of messages from the CAN bus of an electric vehicle and GPS traces from four different drivers travelling on the same route, obtaining an accuracy of over 95% in the identification rate. Furthermore, our evaluation shows that UBM schemes outperform classification approaches traditionally adopted in driver identification literature by a significant margin.

Keywords: Public Transport, Network Modelling
Abstract: Implementing dedicated bus lanes on a network can improve bus travel times at the detriment of cars due to the reduced car capacities, especially at intersections. The optimum solution for the system considering both car and bus passengers is often unclear. In this paper, a bi-level optimization program is developed to optimize the location of dedicated bus lanes on a network while accounting for dynamic queue development at signalized intersections and possible queue spillbacks. The lower level evaluates the travel times on the network and the upper level determines the combination of dedicated bus lane locations that will minimize the objective function. The proposed method is applied to a test network and tested for both saturated and under-saturated traffic conditions with uniform and nonuniform demand patterns. Results show that, the proposed methodology successfully accounts for the effects of queuing. For under-saturated conditions, decrease in travel time of buses outweighs the increase in car travel time when dedicated bus lanes are implemented and the optimum solution finds that bus lanes should be implemented everywhere. For saturated conditions, an optimum solution where bus lanes are implemented only on the periphery of the network is found.

Keywords: Public Transport, Genetic Algorithm, Decision Modeling
Abstract: Bus operators plan the dispatching times of their daily trips based on the average values of their travel times. Given the trip travel time uncertainty though, the performance of the daily operations is different than expected impacting the service regularity and the expected waiting times of passengers at stops. To address this problem, this work develops a model that considers the travel time uncertainty when planning the dispatching times of trips. In addition, it introduces a minimax approach combining Monte Carlo evaluations with a Genetic Algorithm for computing dispatching times which are robust to travel time variations. This approach is tested in a circular bus line of a major bus operator in Asia Pacific (APAC) using 4 months of Automated Vehicle Location (AVL) and Automated Fare Collection (AFC) data for analyzing the travel time uncertainty and computing robust dispatching times. In addition, 1 month of data is used for validation purposes demonstrating a potential service regularity improvement of 5.5% in the average case and 22% in worst-case scenarios.

Keywords: Public Transport, Internet of Things, Vehicle Localization
Abstract: This paper introduces an IoT-based bus location system that can improve the service quality of the bus network and improve the efficiency of operation management. The proposed system provides the bus approach information by utilizing the location and delay information of the bus, and can be realized with low operational cost. The principle of LoRaWAN is used to collect the location information of all buses and deliver the calculated delay information to smart bus stops using electronic paper. A prototype system was implemented, a basic evaluation experiment was conducted and trial calculations for the cost were performed. The corresponding results confirmed that the system can be realized at a lower cost than the bus location system using 3G/LTE.