Keywords: Active and Passive Safety Systems, Driver Assistance Systems, Vehicle/Engine Control
Abstract: This paper investigates the integration of torque vectoring (TV) and active suspension (AS) control on the towing vehicle for yaw rate tracking, and trailer sway reduction in car-semitrailer systems. Due to the nonlinear influence of anti-roll moment distribution, nonlinear model predictive control (NMPC) is employed. Several real-time NMPC algorithms are developed and compared, using either predicted hitch dynamics or estimated hitch joint forces, in a high-fidelity simulation environment with an optimization-based tuning routine. Results show that: (i) integrated AS and TV control outperforms other configurations and, (ii) NMPCs based on hitch joint force estimation are a viable alternative to those embedding hitch dynamics at the cost of reduced robustness

Keywords: X-By Wire Technology, Driver Assistance Systems, Vehicular Signal Processing and Pattern Recognition
Abstract: A key challenge in implementing Steer-by-Wire systems, which omit the mechanical link between steering wheel and front axle, is the dynamic reproduction of road excitation as steering torque feedback for the driver. This research focuses on generating feedback torque using data-based approaches building on measured driving dynamics data from a testing vehicle, as well as simulated data from a multibody simulation. Feedforward and recurrent neural networks are parameterized and trained using previously identified signals, with the resulting rack force and steering torque being compared to reference signals from a conventional steering system. The study utilizes feature generation methods to improve estimation accuracy, and the discussed machine learning estimators are implemented on testing hardware with their computational load being examined for real-time feasibility. The results demonstrate accurate feedback torque estimation of road excitation for both measurement- and simulation-trained neural networks.

Keywords: Vehicular Signal Processing and Pattern Recognition, Multi-Vehicle Systems, Active and Passive Safety Systems
Abstract: Multi-agent trajectory prediction is crucial for autonomous systems. Graph-based methods like GRIP++ provide strong baselines but often use simplistic radial neighborhoods for graph construction and offer opportunities for improvement in graph construction and heterogeneous agent type fusion. This paper introduces Perceptually-Guided Graphs and Type-Aware Fusion (PG-TAF), a suite of enhancements to GRIP++. We propose (1) perceptually-motivated elliptical interaction zones for more realistic fixed graph construction, and (2) a novel mid-fusion 2-stage convolutional strategy for effective agent type-aware feature representation. Comprehensive ablation studies on the ApolloScape dataset demonstrate that these enhancements, particularly their combination, lead to a notable reduction in prediction error, achieving a WSADE of 1.1879 and establishing a new performance benchmark for this line of GRIP++ refinement. Our findings highlight the efficacy of principled graph construction and feature fusion in advancing trajectory prediction.

Keywords: X-By Wire Technology, Vehicle/Engine Control
Abstract: The freedom in designing the feedback torque and gear ratio in a Steer-by-Wire system defines the main benefit to the driver. In this work, the control of the steering wheel system and thus aspects of the steering feel are investigated. The considered control is realized using admittance control, to impose a designed mass-damper feel perceived by the driver. In addition, a feedback torque is provided to the driver based on a dynamic vehicle model. The subsequent velocity control to enable the admittance control is implemented using a model predictive controller. To analyze the frequency response of the system and to provide design guidance, the system is divided into a steering feel and a velocity tracking loop. The velocity tracking loop optimization weights of the linearized model predictive controller provide the desired stability margin and bandwidth for the application if chosen as recommended. Further, the importance of the rejection of the driver's torque in the velocity tracking loop is illustrated using different non-rejected driver models. Knowledge of the driver's torque enables the rejection of this uncertainty, as demonstrated to be a necessity. When designing the damping feel the driver perceives, with the objective of an electric power steering system as a benchmark, the analysis shows a trade-off between the felt damping and the frequencies to which the steering system feels responsive to the driver.

Keywords: Vehicle/Engine Control, Driver Assistance Systems, Navigation and Localization Systems
Abstract: This paper presents the development of a fuzzy logic–based dual adaptive model predictive control (AMPC) framework for robust autonomous vehicle (AV) path tracking under varying road conditions. The framework integrates two AMPCs: one for lateral–yaw control and another for longitudinal control. The lateral controller regulates AV position and yaw, while the longitudinal controller manages the vehicle speed. A fuzzy inference system dynamically adjusts the lateral controller weights, and the reference longitudinal velocity based on specific road condition inputs, including road friction and path curvature. The controllers were designed and implemented in MATLAB and Simulink and evaluated using a standard double-lane-change manoeuvre. The results demonstrate the effectiveness of the proposed framework in achieving integrated trajectory tracking and stability control for AVs.

Keywords: On-Vehicle Sensor Networks, Vehicular Sensor, Vehicle Testing
Abstract: The vehicle sideslip angle is a critical factor in ensuring safe and stable vehicle operation; however, it cannot be measured directly and easily during normal driving. Conventional estimation methods frequently disregard model and measurement uncertainties, offering point estimates that lack a quantitative assessment of their reliability. In order to address these limitations, the present paper proposes a sideslip angle estimator based on a zonotopic Kalman filter. This approach provides both an estimate and a guaranteed bound that accounts for system uncertainties. The approach under consideration explicitly models and propagates measurement noise and parametric uncertainties, resulting in bounded, reliable estimates over time. The validation process is executed through the implementation of co-simulations, employing CarSim, a high-fidelity vehicle dynamics simulator, which is integrated with MATLAB. The findings indicate that the estimated zonotopes consistently envelop the ground truth sideslip angle across a range of driving scenarios, thereby demonstrating the robustness and reliability of the estimator.

Keywords: Vehicle/Engine Control, Vehicular Sensor
Abstract: This paper explores a novel methodology for lateral velocity estimation in autonomous vehicles by combining a Linear Parameter Varying (LPV) Unknown Input Observer (UIO) with a physics-informed neural network (PINN). The ar- chitecture embeds a neural network within the observer frame- work to dynamically estimate time-varying tire parameters, while maintaining physical consistency through constrained learning. The hybrid design enhances robustness against un- measured disturbances and model uncertainties. Real-world experiments conducted on a robotized Renault Zoe demonstrate the effectiveness of the method in improving the accuracy and robustness of lateral speed estimation under both nominal and aggressive driving scenarios.

Keywords: Driver Assistance Systems, Vehicular Signal Processing and Pattern Recognition, Vehicle Testing
Abstract: The growing number of ADAS-equipped vehicles has led to a dramatic increase in driving data, yet most of them capture routine driving behavior. Identifying and understanding safety-critical corner cases within this vast dataset remains a significant challenge. Braking events are particularly indicative of potentially hazardous situations, motivating the central question of our research: Why does a vehicle brake? Existing approaches primarily rely on rule-based heuristics to retrieve target scenarios using predefined condition filters. While effective in simple environments such as highways, these methods lack generalization in complex urban settings. In this paper, we propose a novel framework that leverages Large Language Model (LLM) for scenario understanding and reasoning. Our method bridges the gap between low-level numerical signals and natural language descriptions, enabling LLM to interpret and classify driving scenarios. We propose a dual-path scenario retrieval that supports both category-based search for known scenarios and embedding-based retrieval for unknown Out-of-Distribution (OOD) scenarios. To facilitate evaluation, we curate scenario annotations on the Argoverse 2 Sensor Dataset. Experimental results show that our method outperforms rule-based baselines and generalizes well to OOD scenarios.

Keywords: Driver Assistance Systems, Active and Passive Safety Systems, ICT in Road Safety and Infrastructure
Abstract: Collision detection of real-time autonomous vehicles is a cumbersome task due to the underlying point-cloud computations. To address this, we present an agile and efficient algorithm that is capable of detecting collisions between an ego-vehicle and other obstacles on the road. The proposed algorithm consists of a generalized spherical representation for the objects, which allows for a swift response from the ego-vehicle before an actual collision can occur. Experimental results were simulated using CARLA, Autoware and ROS 2 bridges. They indicate that the proposed approach is at least 200 times faster than Obstacle Collision Checker, a point cloud based algorithm in Autoware. Consistent mean elapsed times for collision detection are observed in the runs and its maximum dispersion is about 1/500textsuperscript{th} of a microsecond.

Keywords: Vehicular Sensor, Image Sensor, Vehicular Signal Processing and Pattern Recognition​
Abstract: Autonomous vehicles (AVs) are expected to revolutionize transportation by improving efficiency and safety. Their success relies on 3D vision systems that effectively sense the environment and detect traffic agents. Among sensors AVs use to create a comprehensive view of surroundings, LiDAR provides high-resolution depth data enabling accurate object detection, safe navigation, and collision avoidance. However, collecting real-world LiDAR data is time-consuming and often affected by noise and sparsity due to adverse weather or sensor limitations. This work applies a denoising diffusion probabilistic model (DDPM), enhanced with novel noise scheduling and time-step embedding techniques to generate high-quality synthetic data for augmentation, thereby improving performance across a range of computer vision tasks, particularly in AV perception. These modifications impact the denoising process and the model's temporal awareness, allowing it to produce more realistic point clouds based on the projection. The proposed method was extensively evaluated under various configurations using the IAMCV and KITTI-360 datasets, with four performance metrics compared against state-of-the-art (SOTA) methods. The results demonstrate the model's superior performance over most existing baselines and its effectiveness in mitigating the effects of noisy and sparse LiDAR data, producing diverse point clouds with rich spatial relationships and structural detail.

Keywords: Active and Passive Safety Systems, Driver Assistance Systems, Vehicle Testing
Abstract: Adaptive Cruise Control has seen significant advancements, with Collaborative Adaptive Cruise Control leveraging Vehicle-to-Vehicle communication to enhance coordination and stability. However, the reliance on stable communication channels limits its reliability. Research on reducing information dependencies in Adaptive Cruise Control systems has remained limited, despite its critical role in mitigating collision risks during sudden braking scenarios. This study proposes a novel fallback longitudinal controller that relies solely on LiDAR-based distance measurements and the velocity of a follower vehicle. The controller is designed to be time-independent, ensuring operation in the presence of sensor delays or synchronization issues. Simulation results demonstrate that the proposed controller enables vehicle-following from standstill and prevents collisions during emergency braking, even under minimal onboard information.

Keywords: Vehicular Sensor, Vehicular Signal Processing and Pattern Recognition, Resilient and Robust Sensing
Abstract: Real-world testing, whether through naturalistic or test-track driving, cannot generate the test coverage required to prove the safety of autonomous vehicles with acceptable statistical significance. Virtual Testing Environments (VTEs) offer a complementary solution by enabling the creation of challenging safety-critical test scenarios that are difficult to replicate in real-world conditions. To ensure that a VTE is a sufficient representation of the real-world, our previous work proposed the development of a VTE using a digital twin of an actual roadway environment. This VTE can generate synthetic LiDAR scans, which can then be compared to real-world scans used to create the digital twin. We employed chamfer distance and density-aware chamfer distance as the metrics for this comparison. In this study, we implemented this approach on a variety of roadway scenes to investigate how different scene compositions affect the comparison metrics. Our findings reveal that factors such as road surface smoothness, vegetation presence, and the LiDAR's point of view (elevation) significantly influence the comparison results.

Keywords: Vehicular Signal Processing and Pattern Recognition
Abstract: In autonomous driving, efficient transmission and precise detection of objects from LiDAR point clouds are essential yet challenging due to high data volume and computational complexity. This paper proposes an ROI-aware octree-based point cloud compression network tailored explicitly for object detection tasks. We introduce a ring-grid module to effectively eliminate redundant ground points, significantly reducing the data size and emphasising regions of interest (ROIs). Subsequently, we apply hierarchical octree encoding to the retained ROI points, utilising a contextual entropy model that integrates ancestor, neighbour, and sibling-child node information. This approach not only achieves a compact representation of the point cloud but also preserves essential semantic context for accurate object detection. Experimental evaluations on the KITTI dataset demonstrate that our method reduces bitrates by over 10% compared to state-of-the-art techniques while simultaneously enhancing object detection precision by approximately 1%, confirming its effectiveness in balancing efficient compression and high detection accuracy.

Keywords: Vehicular Signal Processing and Pattern Recognition, Vehicular Sensor, Driver Assistance Systems
Abstract: This paper presents a multisensor methodology for the detection and global localization of urban poles, using late-stage fusion of high-resolution camera images and LiDAR point clouds. The approach leverages state-of-the-art 2D object detection models, RT-DETR and YOLO11x, which are fine-tuned on semantically annotated data. Detected 2D bounding boxes are projected into 3D space using camera calibration matrices, then filtered, tracked, and globally localized using GPS-IMU fusion and an Unscented Kalman Filter (UKF). Evaluation on a realworld dataset demonstrates a recall of 0.936 and an F1-score of 0.897 for 3D pole global localization. These results highlight the effectiveness of the proposed pipeline for accurate and reliable urban pole mapping, even in the absence of publicly available 3D datasets.

Keywords: Vehicular Signal Processing and Pattern Recognition, Vehicular Sensor, Resilient and Robust Sensing
Abstract: Object detection in an urban environments is essential for autonomous driving, yet collecting and annotating real-world datasets is resource-intensive. This paper presents an automatic annotation pipeline using the CARLA simulator to generate a synthetic dataset of urban trees across various environments and weather conditions. A perception model is then trained for tree detection, and multiple training strategies are employed for systematic evaluation, including pretraining with synthetic data, fine-tuning with real-world data, and sequential fine-tuning. Performance is assessed using precision, recall and F1 score across real-world data. Results show that pretraining with a synthetic dataset substantially reduces the amount of real data required to achieve high performance, with as little as 10% of real data needed to double the F1 score compared to training on synthetic data alone. Furthermore, sequential finetuning, starting from COCO-pretrained weights, followed by synthetic and real data, achieves the highest accuracy. These findings highlight the effectiveness of synthetic data and fine-tuning for efficient and robust urban object detection.

Keywords: Embedded Operating Systems, Telematics, Inter-Vehicular Communication
Abstract: In this paper, we analyze the performance of modern automotive communication middleware and their interactions with the operating system kernel under various operating conditions. Specifically, we examine FastDDS, a widely used open-source middleware, the newly developed Zenoh middleware, and vSomeIP, COVESAs open-source implementation of SOME/IP. Our objective is to identify the best performing middleware for specific operating conditions and understand their interaction with the kernel to achieve best performance. To ensure accessibility, we first provide a concise overview of middleware technologies and their fundamental principles. We first introduce our performance testing methodology designed to systematically assess middleware performance metrics such as scaling performance, end-to-end latency, and discovery times across multiple message types, network topologies, and configurations. Then, we present our methodology to examine the kernel latency contribution and the metrics we consider. Finally, we compare the resulting performance data and present our results in 12 findings.

Keywords: Vehicular Sensor, Resilient and Robust Sensing, Driver Assistance Systems
Abstract: Integrating LiDAR units into passenger vehicles presents a unique set of challenges, primarily due to the larger size of LiDAR compared to other automotive perception sensors. The LiDAR must be packaged in a way that meets multiple vehicle-level requirements, such as aesthetic design, aerodynamic efficiency, manufacturability, ease of assembly, thermal stability, and an unobstructed field of view. Out of these, the design requirement of preserving an aesthetically pleasing vehicle's A-surface has led to a growing demand for LiDAR units to be concealed behind secondary windows. However, this placement could potentially degrade LiDAR performance by introducing unintended effects such as optical distortions, signal attenuation, and reflection artifacts. This preliminary study investigates the impact of placing a LiDAR sensor behind a secondary window on sensor performance, with the primary objective of developing a robust and replicable test methodology. Using a state-of-the-art LiDAR unit and a secondary window sample, data was collected across various target types and analyzed through point cloud comparisons. Although the initial results do not show evidence of significant performance degradation, further testing is required to fully characterize the influence of secondary windows under different configurations and test conditions.

Keywords: Navigation and Localization Systems, Image Sensor, Vehicular Sensor
Abstract: Extrinsic calibration represents a paramount problem in modern automated vehicles. This paper describes a calibration pipeline for LiDAR systems. A trajectory-based approach that relies on LiDAR odometry estimates the relative pose of each sensor pair to provide an accurate initial guess for a visual refinement step. An observability-based filter marginalizes out non-informative trajectory segments through a sensitivity index definition, which potentially introduces noise and inaccuracies into the problem formulation. By excluding these segments, which are prone to introducing noise and inaccuracies, the estimator is guided to focus exclusively on the most informative portions of the sensor trajectories, which exploits by registrating the pointcloud data. The proposed architecture has been extensively tested both with synthetic and real-world data.

Keywords: Embedded Operating Systems, Vehicle Testing
Abstract: This paper presents a modular, open-source software framework for real-time Controller Area Network (CAN) message handling and visualization for electrified off-road vehicles. The framework has been validated on an electrified compact track loader with both electric and hydraulic components arranged in the Hybrid Hydraulic-Electric Architecture(HHEA) to increase energy efficiency and reduce the size of electric motors. Designed to support SAE J1939 and other protocols via DBC-based configuration, the framework integrates a JSON-driven finite state machine (FSM) for condition-based transitions, safety interlocks, and fault handling. It is validated through Hardware-in-the-Loop (HIL) testing and field deployment, enabling seamless communication among subsystems, such as the Battery Management System (BMS), Power Distribution Unit (PDU), inverters, and Vehicle Interface Gateway (VIG). A custom Qt-based dashboard supports real-time signal monitoring and diagnostics. The framework’s extensibility, protocol abstraction, and open-source availability make it suitable for reuse across electrified off-road and heavy-duty vehicle platforms.

Keywords: Heavy Vehicles and Freight Traffic Data Collection, Analytics, and Traffic Control, Energy Consumption, Vehicular Signal Processing and Pattern Recognition
Abstract: Energy transition in transportation sector is experiencing a rapid growth. Although massive renewals of outdated vehicle fleet with Electric Vehicles (EVs), battery management remains one of its main weaknesses. For this reason, research in this field is increasingly expanding, with the target of developing technologies able to combine improved performance with reduced environmental impact. This paper explores the challenging application of Machine Learning (ML) techniques for estimating the battery State of Charge (SoC) without relying on the electrical input data, typically required by classical model-based approaches. Following the set-up and optimization phases of ML models being able to associate vehicle speed to battery SoC, discharging profiles were extrapolated and analyzed to estimate the Remaining Useful Life (RUL). The proposed ML models achieved accurate results, with performance comparable to a traditional observer, despite relying on fewer and simpler variables. These results highlight the potentialities of data-driven approaches for realizing a scalable and easier battery monitoring.

Keywords: Energy Consumption, Vehicular Power Networks, Vehicle/Engine Control
Abstract: Decarbonizing the traffic sector requires public transport to switch to alternative drive technologies. Despite the increasing spread of electric vehicles, there are still challenges in the electrification of public transport. These challenges are particularly pronounced for electrified 18 m-buses. Due to the limited range of 18 m-buses, they are not suitable as a direct replacement for their diesel equivalents, making scheduling difficult for bus operators. A promising approach to extending the range is the integration of an additional energy source on board. This approach is implemented in fuel cell range extender (FC-REX) buses. It combines a battery and a fuel cell power supply, whereby the energy is mainly provided by the battery. The power distribution of the two energy sources is controlled by an energy management system (EMS). In this paper, a Dynamic Programming (DP) approach is proposed to determine the optimal fuel cell load in the EMS using a simulation of an FC-REX bus. The aim of this approach is to find out whether the energy efficiency can be improved and thus the range can be increased. To enable a real-time capable EMS without prior knowledge of the route, a neural network is investigated to approximate the DP control by learning optimal control based on DP data.

Keywords: Vehicular Signal Processing and Pattern Recognition, Vehicular Power Networks, Embedded Operating Systems
Abstract: Due to increasing safety demands for automated driving, automotive electronic control units (ECUs) need to ensure safety-critical functionality even in fault cases. Normally, ECUs feature voltage stabilization circuits to minimize voltage fluctuations in case of power supply disturbances. Especially, large electrolytic capacitors are used as local energy storage for stabilization. These capacitors are subject to aging and may lose their capacitance or fail completely. Currently, the degradation of the capacitors due to aging is not monitored which can heavily influence the fail-operational behavior of the systems. In this contribution, the focus is on a diagnosis concept for the input stabilizing capacitors of automotive ECUs. For characterization, when a DC voltage is applied by electronic fuses, the initial charging process is measured. Based on the measured voltages and currents during the charging process, a data-driven online diagnostic approach is implemented on a microcontroller to estimate capacitance and equivalent series resistance of the capacitor. In an exemplary power supply system, training, validation and test data is generated. The implemented algorithm shows maximum deviations of 18 % for the capacitance and equivalent series resistance estimation of the electrolytic capacitors.

Keywords: Energy Consumption, Heavy Vehicles and Freight Traffic Data Collection, Analytics, and Traffic Control, Multi-Vehicle Systems
Abstract: Heavy-duty trucks consist of a truck unit and a semi-trailer. In addition to the truck unit, the semi-trailers can be made electric with an electrical energy storage (EES) and an electric axle (e-axle). The resulting system is called an e-trailer, which can be combined with any kind of truck unit. When an e-trailer is combined with an internal combustion engine (ICE) truck unit, the resulting system functions as a hybrid vehicle. This configuration offers reduced fuel consumption, which also lowers greenhouse gas emissions and fuel costs. Alternatively, combining an e-trailer with an electric truck unit increases the truck's capacity and driving range. This allows longer haulages and can reduce infrastructure needed for charging the truck’s batteries. The system has some drawbacks, such as the initial cost of the e-trailer and the increased mass. This study evaluates the benefits of an e-trailer compared to a regular trailer with both an ICE and electric truck unit

Keywords: Energy Consumption, Vehicular Sensor, Vehicular Measurement Technology
Abstract: This paper presents an experimental investigation and performance evaluation of a hybrid electric radio-controlled car powered by a Nickel-Metal Hydride battery combined with a renewable Proton Exchange Membrane Fuel Cell system. The study evaluates the performance of the system under various load-carrying scenarios and varying environmental conditions, simulating real-world operating conditions including throttle operation. In order to build a predictive model, gather operational insights, and detect anomalies, data-driven analyses using signal processing and modern machine learning techniques were employed. Specifically, machine learning techniques were used to distinguish throttle levels with high precision based on the operational data. Anomaly and change point detection methods enhanced voltage stability, resulting in fewer critical faults in the hybrid system compared to battery-only operation. Temporal Convolutional Networks were effectively employed to predict voltage behavior, demonstrating potential for use in planning the locations of fueling or charging stations. Moreover, integration with a solar-powered electrolyzer confirmed the system’s potential for off-grid, renewable hydrogen use. The results indicate that integrating a Proton Exchange Membrane Fuel Cell with Nickel-Metal Hydride batteries significantly improves electrical performance and reliability for small electric vehicles, and these findings can be a potential baseline for scaling up to larger vehicle

Keywords: Vehicular Sensor, Resilient and Robust Sensing, Image Sensor
Abstract: Accurate 3D object detection is essential for autonomous vehicles to navigate safely in complex real-world environments. Bird’s Eye View (BEV) representations, which project multi-sensor data into a top-down spatial format, have emerged as a powerful approach for robust perception. While BEV-based fusion architectures have demonstrated strong performance through multimodal integration, the impact of sensor occlusions, caused by adverse environmental conditions such as fog, haze, or physical obstructions, remains underexplored. In this work, we investigate how occlusions affect both camera and LiDAR data using the BEVFusion architecture, evaluated on the nuScenes dataset. We assess detection performance using mean Average Precision (mAP) and the nuScenes Detection Score (NDS). Results show that moderate camera occlusions lead to a 41.3% mAP drop (from 35.6% to 20.9%) in camera-only settings. LiDAR-only detection is more robust but suffers a 47.3% mAP drop under heavy occlusion (from 64.7% to 34.1%), especially for long-range objects. In multimodal fusion, camera occlusion causes a minor 4.1% mAP reduction (to 65.7%), while LiDAR occlusion results in a 26.8% drop (to 50.1%). These findings highlight the BEVFusion model’s stronger reliance on LiDAR and underscore the need for occlusion-aware evaluation and more resilient sensor fusion strategies.

Keywords: Navigation and Localization Systems, Driver Assistance Systems
Abstract: Ensuring safety in autonomous driving requires continuous map maintenance supported by reliable quality indicators. In this context, it is crucial to identify when and where map updates should be triggered, for instance through crowdsourced data, and under which conditions a new map compilation should be deployed. This paper focuses on effective metrics for assessing the quality of vector maps and guiding such decisions. We present a new metric called GOSPAM designed to measure map discrepancies in terms of location errors, existence, and completeness. Through detailed simulations on both point and polyline feature maps, we analyze its sensitivity to common map degradation such as bias, false positives, false negatives, and coordinate errors. The results demonstrate that GOSPAM offers a unified and interpretable measure that effectively captures various forms of map deviation, making it a strong candidate for map quality assessment in automotive applications.

Keywords: ICT in Road Safety and Infrastructure
Abstract: Urban intersections often struggle to balance traffic efficiency with the safety of Vulnerable Road Users (VRUs) such as pedestrians. While traditional traffic light control (TLC) methods focus on optimizing vehicle flow, they often neglect pedestrian safety. Therefore, we propose SafeCrossLight, a deep reinforcement learning (DRL) - based approach that aims to address both efficiency and safety in a unified framework. By factoring pedestrians’ safety into the learning process, SafeCrossLight enables a more responsible and adaptive decision-making process at intersections. Our method is evaluated through Simulation of Urban Mobility (SUMO) and compared with several state-of-the-art TLC approaches. Results show that SafeCrossLight significantly reduces unsafe pedestrian behaviors while maintaining strong efficiency in both vehicle and pedestrian flow, suggesting its high potential for real-world deployment in urban traffic systems.

Keywords: Driver Assistance Systems, Vehicular Signal Processing and Pattern Recognition, Energy Consumption
Abstract: Abstract—Achieving real time performance is an important goal for automated driving and ADAS applications. One optimisation for such systems is the use of lower resolution images for CNN based object detection, which can greatly improve inference speed. Reducing image resolution reduces the size of the image but also reduces image quality. The downsampling method used in ADAS is a topic often not considered when downsizing images, and this study aims to address this gap. This study investigates how downsampling using different interpolation methods impacts machine vision performance. Several common machine vision algorithms are trained on downsampled data, and their performance is evaluated. The downsampling methods used are: Bilinear interpolation, Bicubic interpolation, Nearest Neighbour interpolation, Area-Based resampling and Lanczos4 interpolation. The results show that training with different downsampling methods does have a consistent impact on performance across different object detection algorithms; however, the differences are generally very small, with a difference of less than 2% AP50 in most cases. One object detection model (RT-DETR) is shown to be much more sensitive to interpolation methods. This study indicates which methods of downsampling are best suited for use in ADAS applications, and their relative advantages and disadvantages of each method. The results presented here are relevant to designers of ADAS who are concerned with real-time optimisations.

Keywords: Vehicular Signal Processing and Pattern Recognition, Image Sensor, Driver Assistance Systems
Abstract: Pedestrian segmentation in automotive perception systems faces critical safety challenges due to metamerism in RGB imaging, where pedestrians and backgrounds appear visually indistinguishable. This study investigates the potential of hyperspectral imaging (HSI) for enhanced pedestrian segmentation in urban driving scenarios using the Hyperspectral City v2 (H-City) dataset. We compared standard RGB against two dimensionality-reduction approaches by converting 128-channel HSI data into three-channel representations: Principal Component Analysis (PCA) and optimal band selection using Contrast Signal-to-Noise Ratio with Joint Mutual Information Maximization (CSNR-JMIM). Three semantic segmentation models were evaluated: U-Net, DeepLabV3+, and SegFormer. CSNR-JMIM consistently outperformed RGB with an average improvement of 1.44% in Intersection over Union (IoU) and 2.18% in F1-score for pedestrian segmentation. Rider segmentation showed similar gains with 1.43% IoU and 2.25% F1-score improvements. These improved performance results from the enhanced spectral discrimination of optimally selected HSI bands, effectively reducing false positives. This study demonstrates robust pedestrian segmentation through optimal HSI band selection, showing significant potential for safety-critical automotive applications.

Keywords: Vehicular Signal Processing and Pattern Recognition, Active and Passive Safety Systems, Driver Assistance Systems
Abstract: Accurate prediction of pedestrian crossing intention is a critical requirement for the safety of autonomous vehicles. This paper introduces MSLSTM-PID, a Multi-Stream LSTM framework that sets a new standard for Pedestrian Intention Detection. Our innovative approach lies in its ability to integrate diverse data modalities including kinematics, pose, and egovehicle dynamics with a novel stream of engineered panoptic scene features. Each modality is processed by a dedicated bidirectional LSTM with temporal self-attention, and an adaptive fusion layer dynamically prioritizes the most salient information. To further enhance model robustness, we introduce a targeted Gaussian noise augmentation strategy during training. Comprehensive evaluations on the PIE and JAAD datasets demonstrate that our framework not only achieves state-of-the-art performance with an F1-score of 0.8992 but also exhibits exceptional generalization to unseen driving scenes, affirming its potential for real-world deployment.

Keywords: Driver Assistance Systems, Road Accident Investigation and Risk Assessment, Vehicle Testing
Abstract: For the verification and validation of Automated Driving System features, vehicle manufacturers use scenario-based approaches to break down the world into more manageable traffic scenarios. Since this approach is typically independent from systems engineering processes such as requirements engineering, the Scenario as Specification approach was proposed, which unifies both approaches by using scenarios as a central development artifact. This work focuses on the integration of a technology-independent and technology-dependent safety engineering to the Scenario as Specification approach.

Therefore, we present (1) an extension of the Scenario as Specification approach by introducing a technology-independent feature-level and a technology-dependent system-level view, (2) a deeper integration of Functional Safety and Safety Of The Intended Functionality, (3) the corresponding adaptations of the scenario structure, and (4) an end-to-end practical example with a focus on safety.

Our work aims to lay the foundation for a unified scenario-based approach in industry and research communities for the development and deployment of Automated Driving System features.

Keywords: Heavy Vehicles and Freight Traffic Data Collection, Analytics, and Traffic Control, Active and Passive Safety Systems, Driver Assistance Systems
Abstract: With the rapid advancement and commercialization of autonomous driving technologies, the establishment of safety driving policies has become increasingly imperative. In response to emerging challenges, Mobileye's Responsibility-Sensitive Safety (RSS) model presents a mathematical framework. For its application to heavy-duty vehicles, the varying cargo load that can lead to unstable behavior should be considered when developing the safe autonomous systems for heavy-duty vehicles. In this paper, the distance control strategy for autonomous heavy-duty vehicles considering cargo load is proposed based on the RSS model. The formulation for minimum braking deceleration is derived according to the cargo load and speed through the simulation experiments. The proposed strategy was validated using SIMULINK and dSPACE's ASM (Autonomous Simulation Models) and Adaptive Cruise Control (ACC) was used to track desired clearance in real-time with derived equation for minimum braking deceleration. When the RSS model did not account for cargo load, a collision with the front vehicle occurred under emergency braking scenario. However, the collision can be avoided when the proposed strategy is applied by considering the cargo load.

Keywords: Driver Assistance Systems, Vehicle/Engine Control, Active and Passive Safety Systems
Abstract: Emergency steering maneuvers offer significant safety benefits by enabling the navigation of perilous traffic situations. Extensive research has been conducted to develop emergency maneuvering systems for passenger vehicles. However, performing evasive maneuvers with semi-trailer trucks presents unique challenges that limit their maneuvering capabilities, such as restrictions on the power steering system. This paper presents a unified control approach for automatic evasive maneuvering that integrates steering and differential braking. Differential braking, applied within conservative limits, is used to generate an additional yaw moment to complement the steering, thus enhancing the lateral dexterity of the vehicle. A jerk-optimal quintic polynomial path generation approach is adopted. For path tracking, a linear-quadratic controller (LQR) is employed. The proposed system is evaluated via simulations in MATLAB/Simulink. The simulation results show a significant improvement in evasive performance based on our assessment criteria, the Last Point to Brake (LPTB) and the Last Point to Steer (LPTS).

Keywords: Active and Passive Safety Systems, Vehicle Testing, Multi-Vehicle Systems
Abstract: As industrial autonomous ground vehicles are increasingly deployed in safety critical environments, ensuring their safe operation under diverse conditions is paramount. This paper presents a novel approach for their safety verification based on systematic situation extraction, probabilistic modelling, and verification. We build upon the concept of a situation coverage grid, which exhaustively enumerates environmental configurations relevant to the vehicle's operation. This grid is augmented with quantitative probabilistic data collected from situation-based system testing, capturing probabilistic transitions between situations. We then generate a probabilistic model that encodes the dynamics of both normal and unsafe system behavior. Safety properties extracted from hazard analysis and formalised in temporal logic are verified through probabilistic model checking against this model. The results demonstrate that our approach effectively identifies high-risk situations, provides quantitative safety guarantees, and supports compliance with regulatory standards, thereby paving the way for robust deployment of autonomous systems in real-world domains.

Keywords: Active and Passive Safety Systems, Embedded Operating Systems, System-on-a-Chip
Abstract: Ensuring the functional safety of motion planning modules in autonomous vehicles remains a critical challenge, especially when dealing with complex or learning-based software. Online verification has emerged as a promising approach to monitor such systems at runtime, yet its integration into embedded real-time environments remains limited. This work presents a safeguarding concept for motion planning that extends prior approaches by introducing a Time Safeguard. While existing methods focus on geometric and dynamic feasibility, our approach additionally monitors the temporal consistency of planning outputs to ensure timely system response. A prototypical implementation on a real-time operating system evaluates trajectory candidates using constraint-based feasibility checks and cost-based plausibility metrics. Preliminary results show that the safeguarding module operates within real-time bounds and effectively detects unsafe trajectories. However, the full integration of the Time Safeguard logic and fallback strategies is ongoing. This study contributes a modular and extensible framework for runtime trajectory verification and highlights key aspects for deployment on automotive-grade hardware. Future work includes completing the safeguarding logic and validating its effectiveness through hardware-in-the-loop simulations and vehicle-based testing. The code is available at: https://github.com/TUM-AVS/motion-planning-safeguard

Keywords: Telematics, On-Vehicle Sensor Networks, Vehicle Testing
Abstract: Although prior works reported vulnerabilities introduced in vehicle diagnostics, finding specific service disruption time intervals through diagnostic sessions of protocols such as the Unified Diagnostic Services (UDS) remains unaddressed. In this paper, we report that sending a large volume of Extended Diagnostic Session (EDS) requests to an Electronic Control Unit (ECU) of a passenger car can lead to a Denial-of-Service (DoS) situation. We demonstrated our findings on the Body Control Module (BCM) ECU of a 2015 SUV model. The attacks were conducted using a prototype dongle attached to the OBD-II port, primarily designed for diagnostic purposes. Our findings highlight that sending a large volume of EDS requests at 0.6 ms and 1.0 ms time intervals can result in a DoS situation, which might compromise crucial tasks of vehicular diagnostics, ranging from monitoring I/O functions to ECU firmware updates.

Keywords: Inter-Vehicular Communication, Multi-Vehicle Systems, Telematics
Abstract: The rapid advancement and deployment of connected and autonomous vehicles (CAVs) present transformative opportunities to enhance safety, efficiency, and convenience within the transportation industry. However, these innovations introduce significant cybersecurity risks due to the complex electronics and continuous connectivity that CAVs depend on. Traditional testing methods, while critical, often fall short in detecting vulnerabilities across the vast range of scenarios these vehicles may encounter. Formal verification, a mathematical approach to system validation, offers a more rigorous and comprehensive solution by ensuring that systems operate as expected to search through all possible execution paths. However, defining appropriate system properties for verification remains a challenge, as a system designer may write properties that fail to address real-world threats effectively. This research addresses this gap by integrating threat analysis into the process of defining security properties, ensuring that the verification process is aligned with actual cybersecurity risks. We leverage Natural Language Processing (NLP) to extract key security details from threat analysis result texts, automating the generation of system properties. This approach simplifies the verification process, with its usability demonstrated through a high-level 5G-V2X design use case scenario.

Keywords: Active and Passive Safety Systems, Vehicle Testing
Abstract: The introduction of highly automated vehicles (HAVs) will allow vehicle occupants to take advantage of new seating configurations, such as sitting rearward in the first row. One critical aspect of assessing occupant safety during high-speed impacts is the lack of a dedicated safety framework for rearward-facing passengers in the first row. This paper introduces a method to develop new assessment criteria for these novel seat configurations. Thus, this research presents some preliminary results of rearward-facing occupant injury biomechanics analyses carried out employing a variety of anthropomorphic test devices (ATDs) and the VIVA+ 50M human body model (HBM), restrained with different belt configurations and considering different seat typologies. It reviews the suitability of 50th percentile male ATDs to capture a biofidelic engagement with the seat structure and belt system and evaluates the reaction loads on the occupant, along with the energy management resulting from seat back rotational stiffness and energy-absorbing foams layered behind the seat cushion. Based on the results, the THOR-AV-50M is a suitable candidate for further biofidelity analysis. Torso occupant loads can be effectively reduced utilizing seat back rotation but pelvis load management requires further studies.

Keywords: Image Sensor, System-on-a-Chip, Vehicular Sensor
Abstract: The passenger information system in railway sector is essential for helping passengers with schedules, connectivity, and overall comfort. It delivers on-board passenger information for trams, trains, and buses as an innovative solution with real-time information. Our proposed method utilizes an edge computing based computer vision algorithm to accurately count passenger data on an embedded computing board, including the number of passengers entering and exiting, as well as the current load in each carriage. This information is then presented in a way that helps passengers choose the carriage or route that best maintains their comfort. The system's design and development, from the sensor device to the API specification and its visual representation, are developed as part of a comprehensive product development process. The proposed method tested with an accuracy of 98% using data collected from a wide angle TOF camera-based counting system that utilized passenger detection and tracking.

Keywords: Navigation and Localization Systems, Inter-Vehicular Communication, Vehicle/Engine Control
Abstract: This paper presents a simulation system for autonomous delivery vehicles that supports real-time reception of optimal routes from a supervisory control system and performs corresponding path-following simulations. The system consists of a vehicle dynamics model, a realistic road environment, and an Ethernet-based communication framework for real-time data exchange. A path generation module was implemented to reconstruct routes based on received GPS coordinates, while a lateral control model using Pure Pursuit logic ensured accurate trajectory tracking. Simulation results confirm that the proposed system enables reliable real-time route tracking, validating its effectiveness for supervisory-based autonomous delivery applications.

Keywords: Energy Consumption, Driver Assistance Systems, Active and Passive Safety Systems
Abstract: Nighttime headlamp usage significantly contributes to vehicle energy consumption and can cause unnecessary glare when operated continuously at full intensity. In this paper, we propose a self-learning Random Forest framework designed to optimize headlamp power consumption under low-light driving conditions. We developed a balanced synthetic dataset composed of predefined, labeled use cases simulating diverse nighttime scenarios. To effectively increase our training data without additional manual annotation, we adopted a semi-supervised self-training approach, iteratively adding the model's high-confidence predictions on unlabeled samples back into the training set. Additionally, specialized submodels were trained to resolve confusion between certain pairs of labels representing closely related nighttime driving scenarios (e.g., distinguishing between moderate dimming with and without beam narrowing), achieving an average accuracy improvement of approximately 2.7% over the base model alone. Experimental results on synthetic datasets and real-world nighttime camera feeds demonstrate that this ensemble approach consistently outperforms the base model, particularly in differentiating similar lighting scenarios. Furthermore, our framework acts as a semi-automated annotator by flagging uncertain label transitions, significantly speeding up the annotation of real-world datasets, thus enabling faster availability of labeled nighttime driving data for various automotive applications.

Keywords: Multi-Vehicle Systems
Abstract: Traditional fixed-route public transit systems often struggle with inefficiency in suburban areas characterized by low population density and fluctuating demand. This paper presents and validates DynaRyde, a framework for optimizing suburban public transit by integrating dynamic route scheduling with on-demand passenger requests using well-established algorithms. Our approach utilizes the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm to group spatially and temporally proximate ride requests, generating “virtual stops” in real time. Routes are then dynamically computed using a Traveling Salesman Problem (TSP)-like optimization heuristic, incorporating both mandatory transit hubs and virtual stops. We implemented and evaluated the DynaRyde system using the Simulation of Urban MObility (SUMO) traffic simulation environment on a realistic suburban road network under various demand scenarios. Performance was benchmarked against an enhanced fixed-route system and a non-clustering on-demand model. Results demonstrate that our integrated system significantly improves service quality and efficiency, achieving up to a 19% increase in service coverage and a 15-35% reduction in average waiting times compared to a resource-equivalent fixed-route baseline. These findings highlight the substantial potential of integrating on-demand clustering and dynamic routing strategies to create more adaptable and passenger-centric public transit solutions.

Keywords: ICT in Road Safety and Infrastructure, Resilient and Robust Sensing, Image Sensor
Abstract: RGB and event cameras, as commonly used vision-based sensors in intelligent transportation systems (ITS), have been widely applied in traffic surveillance. Vision-based roadside traffic monitoring datasets provide a strong foundation for training the perception models in robust traffic surveillance systems. However, most existing datasets are limited to a single sensor modality or a single viewpoint. To address this gap, we present the TUMTraf CrossVision dataset, a multi-view multi-modal vision dataset that captures data at an arterial intersection in the Garching-Hochbrück industrial area of Munich, Germany, using 4 pairs of temporally synchronized RGB and event cameras, each covering one direction of the intersection. The dataset comprises 23,128 images collected over two months under varying illumination conditions (day, twilight, and night) and diverse weather conditions (clear and rainy), along with 83,026 precisely annotated 2D bounding boxes for 6 common traffic participant classes in YOLO and COCO formats. We evaluated TUMTraf CrossVision with the state-of-the-art object detection approach on both RGB and event data, demonstrating its value as a resource for future research on intelligent roadside traffic surveillance systems. The dataset will be publicly available at: https://innovation-mobility.com/tumtraf-dataset.

Keywords: Navigation and Localization Systems, Vehicular Signal Processing and Pattern Recognition, Image Sensor
Abstract: In this paper, we study a method of acquiring sign information for the purpose of determining safe routes in a bicycle navigation system. The proposed methods of acquiring sign information are based on Google Street View and a method using a drive recorder. Yolov8, a real-time object detector, is used for image recognition. In preliminary experiments, we evaluated the recognition accuracy of signs from drive recorder videos based on driving videos on motorways. As a result, it was determined that the recognition accuracy using the Yolov8 model with 100 training cycles was the most suitable, and a comparison of the two methods was conducted using this weight file. The comparison results showed that both methods have the same recall (78%), but the precision is 41% for the conventional method and 19% for the proposed method, indicating that the use of a drive recorder often results in false positives.

Keywords: Road Accident Investigation and Risk Assessment, ICT in Road Safety and Infrastructure, Navigation and Localization Systems
Abstract: The traditional route optimization policies of the urban networks are focused on distance or temporal efficiency without considering the dynamic safety risks of accident-prone locations, which exposes motorists to high risks. The paper analyzes a technique that minimizes the travelling time and detects risk-aversive paths on urban road networks. The analysis uses official UK government traffic and accident data from 2015-2024 to estimate regional accident risk indices and combines these in a graph optimization framework based on OpenStreetMap (OSM) to produce routes under differing safety-time trade-offs. Findings indicate a strong positive correlation between growing traffic levels and accident risk indices. In high-risk areas like Kent, risk-sensitive routing algorithms produce significantly safer travel suggestions compared to traditional shortest path algorithms, while low-risk areas show no significant variations between applications. These results explain why risk-sensitive routing is important in urban settings and serve as a practical foundation for studying next-generation navigation systems and advanced driver assistance systems (ADAS). The study acknowledges shortcomings including lack of temporal, weather and intersection-level details, suggesting future studies should include real-time environmental data.