Browse Topic: Autonomous vehicles

Items (3,078)
To address the performance testing requirements of autonomous vehicles (AVs), this study proposes a model predictive control (MPC) algorithm specifically designed for low-ground-clearance test target vehicles (TTVs) to achieve trajectory tracking control. First, the kinematic model of the TTV is established, and its state-space equations are derived. An objective optimization function incorporating both error weighting and control weighting is designed. Simulation analysis reveals the influence of the control error weighting ratio (CEWR) on both straight-line and curved trajectory tracking performance: For straight-line tracking, increasing the CEWR from 10 to 25 reduces the overshoot, but increases the distance required to reach the target trajectory by 4.7%. A similar pattern is observed in curved trajectory tracking. To overcome the limitations of the fixed CEWR, an improved MPC algorithm integrating fuzzy control is proposed. This algorithm dynamically adjusts the CEWR in real time
Ji, ShaoboLu, YueqiLiao, GuoliangChen, ZhongyanLi, MengLyu, ChengjuZhang, Zhipeng
This paper presents a structured test plan for the development and validation of a Self-Propelled Trailer (SPT), an emerging concept designed to enhance the towing capacity of compact, fuel-efficient vehicles. Unlike conventional trailers, the proposed system integrates electric propulsion and autonomous sensing to actively assist the towing vehicle, reducing engine load and improving both safety and fuel economy. The methodology employs a Design Failure Mode and Effects Analysis (DFMEA) to systematically identify potential risks, while incorporating Society of Automotive Engineers (SAE) standards to guide environmental durability testing (dust, water ingress, gravel impact) and dynamic performance evaluations (gradeability, braking, and stability). A comprehensive set of test procedures is outlined to validate system reliability, robustness, and compliance with established towing requirements. The study demonstrates how powered trailer technology can extend the practical use of
Reilly, CarterPeters, DianeZadeh, Mehrdad
The concept of the vehicle has changed as a result of many innovations over the last decade in the fields of connected, autonomous/automated, shared, and electric (CASE) technologies. At the same time, labor shortages in Japan are becoming more serious due to a decline in the working population. To help resolve these issues, a remote-controlled autonomous vehicle driving system called Telemotion has been developed that automates the movement of vehicles in production plants. This system is an autonomous driving and transportation system in which the recognition, judgment, and operation functions of driving are handled by a control system outside the vehicle that communicates wirelessly with the vehicle. This system utilizes artificial intelligence (AI) and other advanced technologies to realize safe unmanned autonomous driving, and is already in operation in production plants. Currently, efforts are under way to build a digital twin environment and conduct AI learning using computer
Hatano, YasuyoshiIwazaki, NoritsuguNagafuchi, YuheiIwahori, KentoTanaka, AtsushiUezu, SatoruKanou, TakeshiInoue, GoOkamoto, YukiOka, YuheiKakuma, DaisukeChiba, HiroyaEgashira, KazukiIshikuro, MegumiSawano, Takuro
With the steady increase in autonomous driving (AD) and advanced driver-assistance systems (ADAS) aimed at improving road safety and navigation efficiency, simulation tools have become a critical part of the development process, allowing systems to be tested while mitigating the risk of physical injury or property damage upon failure. Physics-based simulators are central to virtual vehicle development, yet their control responses often differ from real vehicles, potentially limiting the transfer of controllers and algorithms developed in simulation. As these simulations play an important role in the vehicle design and validation process, a critical question is how well their predicted behavior translates to real-world physical systems. This paper presents a calibration framework for an autonomous vehicle platform that learns the motion characteristics of an experimental vehicle and uses that knowledge to correct the actuator response of a simulation model. The model is trained by
Soloiu, ValentinSutton, TimothyMehrzed, ShaenLange, RobinZimmerman, CharlesPeralta Lopez, Guillermo
By the early 2020s, more than 4.5 billion people have been living in urban areas worldwide, compared to just 1 billion in 1960. Rising growth in urban populations present challenges to infrastructure and transportation systems. Higher traffic levels and reliance on conventional vehicles have contributed to heightened greenhouse gas (GHG) emissions, rising global temperatures, and irreversible environmental degradation. In response, emerging transportation solutions—including intelligent ridesharing, autonomous vehicles, zero-tailpipe-emission transport, and urban air mobility—offer opportunities for safer and more sustainable transportation ecosystems. However, their widespread adoption depends not only on technological performance and efficiency, but also on integration with current infrastructure, safety, resilience to unexpected disruptions, and economic viability. A dynamic agent-based System-of-Systems (SoS) transportation model is developed to simulate vehicle traffic and human
Rana, VishvaBalchanos, MichaelMavris, DimitriValenzuela Del Rio, Jose
Autonomous vehicles may attract more passengers to recline their seat for comfort. However, under severe rear-end crashes and large reclining angle, the backward inertia could completely throw occupant out of seat. Even if the occupant body can be restrained by seatbelt, the occupant’s head could slide out of the head restraint area. Any of these situations may cause severe injuries. To address this safety concern, we developed a sliding seat system designed to enhance occupant retention. Activated by impact inertia of rear-end collision, the system allows the seat sliding backward along its track in a controlled manner, and the sliding stroke is accompanied by a restraint force and absorbs some amount of kinetic energy during the sliding. Thus, occupant retention can be enhanced, and injury risks of head and neck can be reduced. To demonstrate this concept, we built a MADYMO model and conducted a parametric analysis. The model includes a 50th percentile human model, a vehicle seat
Dai, RuiZhou, QingPuyuan, TanShen, Wenxuan
This paper contains Part 2 of a two-part paper series proposing potential regulatory approaches for occupant safety in Automated / Autonomous Vehicles (AVs) with unique seating configurations (stagecoach and campfire seating). Part 2 focuses on interior safety sensing, associated messaging, and ride control approaches both prior to and during a ride. Assessments are also proposed after significant vehicle braking and crash events. The proposed conditions are to be assessed in a static vehicle environment with humans segmented by occupant size and an infant dummy. On the vehicle seat and on the vehicle floor occupant detection conditions are proposed along with restraint usage detection conditions for vehicle seat belt usage, Child Restraint Seat (CRS) usage, CRS seat belt usage, and Lower Anchors and Tethers for Children (LATCH) system usage. These conditions may be detected by sensors / computer algorithms and human monitoring and thus are technology agnostic. The topics of animal
Thomas, Scott
Some Automated / Autonomous Vehicles (AVs) have unique seating configurations (stagecoach and campfire seating) which present expanded occupant safety challenges. Significant portions of the National Highway Traffic Safety Administration (NHTSA) Federal Motor Vehicle Safety Standards (FMVSS) do not yet align with AVs containing unique seating. This paper series takes the NHTSA occupant safety standard approach for conventional forward-facing seat vehicles where many compliance evaluations are in the frequently occupied front row and expands it to stagecoach and campfire AVs where the rear seating row is anticipated to be frequently occupied. The approaches proposed are from a logic-based safety-focused analysis and in many cases previously published material. The goal of this paper series is to offer regulatory proposals that enable equivalent performance for these AVs to existing forward-facing seating vehicle occupant safety standards and meet Executive Order 13045 on child safety
Thomas, Scott
Vision-language models (VLMs) are increasingly used in autonomous driving because they combine visual perception with language-based reasoning, supporting more interpretable decision-making, yet their robustness to physical adversarial attacks, especially whether such attacks transfer across different VLM architectures, is not well understood and poses a practical risk when attackers do not know which model a vehicle uses. We address this gap with a systematic cross-architecture study of adversarial transferability in VLM-based driving, evaluating three representative architectures (Dolphins, OmniDrive, and LeapVAD) using physically realizable patches placed on roadside infrastructure in both crosswalk and highway scenarios. Our transfer-matrix evaluation shows high cross-architecture effectiveness, with transfer rates of 73–91% (mean TR = 0.815 for crosswalk and 0.833 for highway) and sustained frame-level manipulation over 64.7–79.4% of the critical decision window even when patches
Fernandez, DavidMohajerAnsari, PedramSalarpour, AmirPese, Mert D.
High-fidelity 3D reconstruction of large-scale urban scenes is critical for autonomous driving perception and simulation. Existing neural rendering methods, including NeRF and Gaussian-based variants, often face challenges like unstable geometry, noisy motion segmentation, and poor performance under sparse viewpoints or varying illumination. This paper presents a self-supervised Gaussian-based framework to address these challenges, enabling robust static–dynamic decomposition and real-time scene reconstruction. The proposed method introduces three innovations: (1) a semantic–geometric feature fusion module that combines semantic context and geometric cues for reliable motion prior estimation; (2) a cross-sequence geometric consistency constraint that enforces depth and surface continuity across time and viewpoints; (3) an efficient Gaussian parameter optimization strategy that stabilizes geometry by jointly constraining scale and normal updates. Experiments on the Waymo Open Dataset
Feng, RunleiWang, NingZhang, Zhihao
Despite remarkable advances in vehicle technology - enhancing comfort, safety, and automation – productivity of transportation over the road continues to decline. Stop-and-go driving remains one of the most persistent inefficiencies in modern mobility systems, leading to greater travel delays, energy waste, emissions, and accident risk. As vehicle volumes rise, these effects compound into systemic challenges, including driver frustration, unstable flow dynamics, and elevated greenhouse gas (GHG) emissions. To address these issues, an extensive data-driven evaluation was performed characterizing the underlying causes of traffic instability and uncovering hidden behavioral parameters influencing traffic flow. This research led to the identification of a previously unrecognized metric - the Driver Comfort Index (DCI) - which quantifies an inter-vehicle spacing behavior that reflects intrinsic human driving behavior. Building on this discovery, mixed traffic is explored to identify its
Schlueter, Georg J.
Ensuring safe operation and reliable control of mobility systems remains a significant challenge, particularly for nonlinear and high-dimensional applications subject to external disturbances with hard constraints and limited computational resources in real-time implementations. A reference governor (RG) can enforce constraints using an add-on scheme that preserves the pre-stabilizing controller while balancing the need to satisfy other requirements, including reference tracking and disturbance rejection. Thus, in this paper, we exploit RG-based strategies focusing on nonlinear mobility systems. While the method is generalizable to other applications, such as waypoint following for autonomous driving, the flight dynamics of a quadrotor system with twelve states are used as an example. We implement a disturbance rejection RG to satisfy safety constraints and track set points. To handle nonlinearity, we propose an optimal strategy to quantify the maximum deviation between the nonlinear
Dong, YilongLi, Huayi
Road grade can impact the energy efficiency, safety, and comfort associated with automated vehicle control systems. Currently, control systems that attempt to compensate for road grade are designed with one of two assumptions. Either the grade is only known once the vehicle is driving over the road segment through proprioception, or complete knowledge of the oncoming road grade is known from a pre-made map. Both assumptions limit the performance of a control system, as not having a preview signal prevents proactive grade compensation, whereas relying only on map data potentially subjects the control system to missing or outdated information. These limits can be avoided by measuring the oncoming grade in real-time using on-board lidar sensors. In this work, we use point returns accumulated during travel to estimate the grade at each waypoint along a path. The estimated grade is defined as the difference in height between the front and rear wheelbase at a given waypoint. Kalman filtering
Schexnaydre, LoganPoovalappil, AmanRobinette, DarrellBos, Jeremy
Advances in Connected and Automated Vehicles (CAVs) have developed a level in which high-definition maps can be used to improve road safety. Data compactness and robustness on road characterization is essential for the proper handling of vehicles under curves. In this paper, an optimization scheme that relates highway-design road curvature and optimal speed of travel is defined to safely navigate through a given road. The scheme is divided in two main steps. First a nonlinear optimization problem, in which curvature profiles are fitted from a model that based on street design standards as per the American Association of State Highway and Transportation Officials (AASHTO). Secondly, the optimized curvature profile is subject to a secondary optimization problem that uses vehicle dynamics for both constraints and objective function derivation. Guidance reference parameters such as curvature and velocity, at different levels of friction are analyzed. Results show that, even in sparse
Jacome, Ricardo OsmarStolle, CodyGrispos, George
LiDAR (Light Detection and Ranging) systems are essential for autonomous driving (AD) and advanced driver-assistance systems (ADAS), providing accurate 3D perception of the surrounding environment. However, their performance significantly deteriorates under adverse weather conditions such as fog, where laser pulses are scattered by airborne particles, resulting in substantial noise and reduced ranging accuracy. This scattering effect makes it difficult to detect objects within or behind particulate matter, posing a serious challenge for reliable perception in real-world driving scenarios. To address this issue, we propose an algorithm that combines adaptive multi-echo signal processing with a feature-integrated, rule-based denoising framework to enhance LiDAR performance in noisy environments. The multi-echo approach selectively utilizes meaningful signal returns by evaluating both intensity and relative echo positions. Based on predefined rules, the algorithm identifies the echo most
Kaito, SeiyaZheng, ShengchaoFujioka, IbukiBeppu, Taro
Off-road autonomous vehicle systems must be able to operate across unstructured and variable terrain while avoiding obstacles. This presents significant challenges in vehicle and control system design, especially for less conventional platforms such as 6×4 vehicles. While forward driving autonomy has developed and matured in recent years, effective reverse navigation remains an under-explored area of vehicle co-design. Reversing 6×4 vehicles have limited rear steering authority, an extended wheelbase, and asymmetric traction, which introduce complex dynamics into any control system that is used. To address this need, a robust and experimentally validated fuzzy logic control architecture for 6×4 reverse navigation was developed during the course of this project. This architecture incorporates both near-field and long-range path data with adaptive outputs controlling steering and velocity based on a rule base that covers the whole vehicle state space. This method has low computational
Dekhterman, Samuel R.Sreenivas, Ramavarapu S.Norris, William R.Patterson, Albert E.Soylemezoglu, AhmetNottage, Dustin
The shared autonomy framework has become an option with great potential in the field of autonomous vehicles. Human and machine control decisions typically demonstrate strengths in different scenarios. As a result, the robustness of systems can be enhanced by the collaboration between humans and autonomy. A shared autonomy architecture that takes into account both human and environmental factors was proposed in this work. The authority distribution between the human operator and the autonomy algorithm was determined by the Shared Autonomy Arbiter (SAB). Designed with a two-tier structure, the SAB incorporated a policy-level decision module, as well as a numerical-level arbitration tuning module. A fuzzy inference system (FIS) was incorporated to enhance the noise tolerance of the policy selection module. Furthermore, the human factor was taken into account by applying a projection to the users’ control input. The human operator’s control decision was projected by the Adaptive
Sang, I-ChenNorris, WilliamPatterson, AlbertSreenivas, Ramavarapu S.Soylemezoglu PhD, AhmetNottage, Dustin S.
With the rise of end-to-end autonomous driving, visual perception for environmental understanding has become a key research topic in advanced driver assistance system (ADAS) development. Most existing end-to-end models generate only executable control commands or planned trajectories, making the prediction process difficult to interpret. In this study, we present an end-to-end approach for traffic-light recognition and stop-sign detection built on top of the open-source openpilot framework. Instead of deploying separate object detection networks, we extend the existing backbone with two lightweight multi-task heads: a traffic-light detection and classification head, and a stop-sign detection head with confidence estimation. The modified architecture preserves openpilot’s core driving functionality by reusing shared features and incorporating compact residual and feed-forward layers. The additional perception outputs are appended to the original outputs, ensuring that the model’s
Wang, HanchenLi, TaozheHajnorouzali, YasamanBurch, Collinli, VictoriaTan, LinArjmanzdadeh, ZibaXu, Bin
Robust perception systems for autonomous vehicles rely heavily on high-quality, labeled data, particularly in off-road and unstructured environments. However, the performance of the perception model is often degraded by data chaos resulting from limitations in automated segmentation. Foundation models, such as SAM2, while powerful, typically generate masks based on low-level visual cues, including color and texture gradients. In complex off-road scenes, this leads to semantic fragmentation. A single object, like a moss-covered log, can be split into not only dozens of segments for its bark and moss but also hundreds of smaller, meaningless patches based on minor color variations. This paper introduces a context-aware annotation agent to resolve this issue. Our workflow integrates a vision-language model (Florence-2) for scene understanding with a segmentation model (SAM2) for mask generation. Instead of segmenting indiscriminately, our agent leverages Florence-2 to comprehend the image
Patil, AshishMikulski, DariuszMwakalonge, JudithJia, Yunyi
This paper presents a comparative study of three widely used cloud platforms, Google Colab, Microsoft Azure, and Amazon Web Services (AWS), for running a real-time cooperative perception system based on roadside unit (RSU) cameras. The goal is to evaluate the performance, scalability, and cost-efficiency of each platform when handling high-volume video data for object detection, a key task in autonomous driving. A unified perception pipeline using the YOLOv8 Small model was deployed on all platforms, with the same dataset and settings to ensure fair comparison. The evaluation focused on key metrics such as latency, frame processing rate, detection accuracy, cost, scalability, and reliability. The results show that Google Colab is a cost-effective starting point but has limitations in uptime and scalability. Azure offers stable performance and balanced cost, making it suitable for medium-scale applications. AWS delivers the best scalability and speed but at a higher cost. This study
Alkharabsheh, EkhlassAlawneh, ShadiRawashdeh, Osamah
Sparse Stream DETR 3D object detection has become pivotal in autonomous driving, and previous methods achieve remarkable performance by aggregating temporal information, which also face a balance problem of precision and efficiency. Knowledge distillation offers a promising solution to enhance the efficiency of a small model without incurring computational overhead; however, previous methods lack the exploration of the Temporal Distillation knowledge for the DETR detector. This paper designs a novel Temporal DETR Query Guidance paradigm to impart temporal relation knowledge from a powerful teacher model to enable the student to associate object states across time, leverage historical context. The teacher’s queries grasp the temporal knowledge through self-attention, and the backbone uses the EVA-02 large-scale image model. The student utilizes the teacher's self-attention layer and its own learnable queries to compute the attention as its guidance and mimics the feature interaction
Yan, Yixiong
Recent years have seen a rapid rise in edge-oriented object detection models, including new YOLO variants and transformer-based RT-DETR. Choosing an appropriate model for vehicle detection, however, remains challenged because common metrics such as precision, recall, and mAP capture only part of the trade-off between accuracy and computational cost. To better support model selection, we introduce the Multi-dimensional Equilibrium Detection Assessment Score (MEDAS), which evaluates detectors across four practical dimensions: performance, balance, efficiency, and adaptability. The framework includes a normalization strategy and adjustable weighting so that evaluations can reflect specific deployment needs, especially in resource-limited settings. Experiments on the MS-COCO vehicle dataset show that while RT-DETR models offer competitive accuracy, they require substantially more computation. In contrast, lightweight YOLO variants provide a stronger balance between accuracy and efficiency
Guo, Bin
Object detection and distance prediction have advanced significantly in recent years. The YOLO toolbox has released its 11th version, along with numerous variants that have been applied across various fields. Meanwhile, the Detection Transformer (DETRs) has repeatedly set new state-of-the-art (SOTA) records in the field of object detection. Depth Anything also released its second version last year, further pushing the boundaries of distance detection. Although these models achieve impressive performance, they often require substantial computational resources. However, for the algorithms intended for real-world applications and deployment on onboard devices, computational efficiency are extremely critical. Inference time per frame is a critical factor in ensuring an algorithm’s reliability and feasibility. Designing a model that operates in real time without sacrificing accuracy remains an extremely challenging problem, and extensive research is ongoing in this area. To address this
Li, TaozheWang, HanchenHajnorouzali, YasamanXu, Bin
Rapidly upcoming deployment of autonomous vehicles (AVs), including robotaxis and trucks, has intensified the need for rigorous safety assessment of complex AI-driven systems. While considerable effort has been invested in constructing safety cases for AVs, systematic approaches for evaluating these safety cases remain underdeveloped. This paper presents a three-stage methodology for assessing AV safety cases. A process for assessing argumentation is presented that involves traceability to pre-reviewed and peer-reviewed safety cases such as the Open Autonomy Safety Case (OASC). Next, we present a structured process for evaluating the quality of evidence supporting these arguments. We applied this methodology to evaluate safety cases from multiple AV developers, enabling iterative refinement throughout the development lifecycle. Our agile approach supports efficient assessments by establishing clear traceability to industry standards and enabling early identification of potential gaps
Wagner, Michael
Ensuring the safety of Vulnerable Road Users (VRUs) is a critical challenge in the development of advanced autonomous driving systems in smart cities. Among vulnerable road users, bicyclists present unique characteristics that make their safety both critical and also manageable. Vehicles often travel at significantly higher relative speeds when interacting with bicyclists as compared to their interactions with pedestrians which makes collision avoidance system design for bicyclist safety more challenging. Yet, bicyclist movements are generally more predictable and governed by clear traffic rules as compared to the sudden and sometimes erratic pedestrian motion, offering opportunities for model-based control strategies. To address bicyclist safety in complex traffic environments, this study proposes and develops a High-Order Control Lyapunov Function–High-Order Control Barrier Function–Quadratic Programming (HOCLF-HOCBF-QP) control framework. Through this framework, CLFs constraints
Chen, HaochongCao, XinchengGuvenc, LeventAksun Guvenc, Bilin
Shared Autonomous Electric Vehicles (SAEVs) can enhance urban mobility and efficiency. However, their operational performance is often hindered by the spatio-temporal imbalance between vehicle supply and passenger demand, leading to long wait times. This paper develops a novel repositioning framework where a lightweight CNN, informed by computationally intensive multi-agent simulations, enables real-time strategy deployment. The results show that: (1) An optimized repositioning policy, calibrated via multi-agent simulation, effectively cuts the mean passenger waiting time from 12.0 to 3.0 minutes (a 75% reduction). (2) A lightweight CNN surrogate model enables real-time deployment, reducing the policy computation time from ~4 hours to ~5 minutes (>98% faster). (3) The deep learning surrogate achieves this speed with a negligible performance trade-off, increasing the waiting time by only 0.156 minutes (4.9%) compared to the full optimization.
Shang, KaiWang, Ning
In order to achieve fully autonomous driving, point to point autonomous navigation is the most important task. Most existing end-to-end models output a short-horizon path which makes the decision process hard to interpret and unreliable at intersections and complex driving scenarios. In this research, we build a navigation-integrated end-to-end path planner on top of an openpilot open source model. We created a navigation branch that encodes route polyline geometry, distance-to-next-maneuver, and high-level instructions and combines with path plan branch using residual blocks and feed-forward layers. By adding minimal parameters, new model keeps the original openpilot tasks unchanged and have the path output based on the navigation information. The model is trained on diverse urban scenes’ intersections, and it shows improved route performance in vehicle testing. The proposed model is validated in a Comma 3x device installed on a 2025 Nissan Leaf test vehicle. The road test results
Wang, HanchenLi, TaozheHajnorouzali, YasamanBurch, Collinli, VictoriaTan, LinArjmanzdadeh, ZibaXu, Bin
This paper presents a scenario generation framework that creates diverse, parametrized, and safety-critical driving situations to validate the safety features of autonomous vehicles in simulation [1]. By modeling factors such as road geometry, traffic participants, environmental conditions, and perception uncertainties, the framework enables repeatable and scalable testing of safety mechanisms, including emergency braking, evasive maneuvers, and vulnerable road user protection. The framework supports both regulatory and edge case scenarios, mapped to hazards and safety goals derived from Hazard Analysis and Risk Assessment (HARA), ensuring traceability to ISO 26262 functional safety requirements and performance limitations. The output from these simulations provides quantitative safety metrics such as time-to-collision, minimum distance, braking and steering performance, and residual collision severity. These metrics enable the systematic evaluation of evasive maneuvering as a safety
Chandra Shekar, KiruthigaArab, Aliasghar
Achieving full vehicle autonomy is not just about adding sensors or compute - it requires a fundamental shift in how vehicles are architected. Autonomous systems rely on higher-resolution sensors, massive processing power, and the ability to fuse data from multiple sources in real time. Centralized in-vehicle architectures, which consolidate computing and enable sensor fusion, place unprecedented demands on connectivity. Precise time synchronization across systems becomes critical, as does advanced control to ensure safe and reliable operation. Any delay or data loss can impact decision-making, making robust, resilient communication links essential. High-performance connectivity is the backbone of this evolution. It must deliver the highest bandwidth to handle massive streams of sensor data, support long-reach connections across the vehicle, and maintain error-free performance even in the most challenging electromagnetic environments. This combination of speed, reach, and reliability
Shwartzberg, Daniel
This paper introduces a novel methodology to enhance the energy efficiency of eco-driving controllers in Connected and Automated Vehicles (CAVs) by leveraging reinforcement learning (RL) techniques for real-time parameter optimization. Traditional eco-driving strategies rely on fixed control parameters, which limit adaptability across diverse traffic and road conditions. To address this, we apply continuous action space RL algorithms, specifically Deep Deterministic Policy Gradient (DDPG) and Proximal Policy Optimization (PPO), to dynamically tune four key parameters within a model predictive control framework that is grounded in Pontryagin’s Maximum Principle (PMP). These parameters influence acceleration, braking, cruising, and intersection-approach behaviors, making them critical for achieving optimal eco-driving performance. Our study employs Argonne National Laboratory’s RoadRunner simulator, a Simulink-based environment designed for high-fidelity CAV analysis, incorporating
Zhang, YaozhongAmmourah, RamiHan, JihunMoawad, AymanShen, DaliangKarbowski, Dominik
Autonomous platforms such as self-driving vehicles, advanced driver-assistance systems (ADAS), and intelligent aerial drones demand real-time video perception systems capable of delivering actionable visual information at ultra-low latency. High-resolution vision pipelines are often hindered by delays introduced at multiple stages—sensor acquisition, video encoding, data transmission, decoding, and display—undermining the responsiveness required for safety-critical decision making. This study introduces a holistic system-level optimization framework that systematically reduces end-to-end video latency while maintaining image fidelity and perception accuracy. The proposed approach integrates hardware-accelerated encoding, zero-copy direct memory access (DMA), lightweight UDP-based RTP transport, and GPU-accelerated decoding into a unified pipeline. By minimizing redundant memory copies and software bottlenecks, the system achieves seamless data flow across hardware and software
Indrakanti, Rama Kiran Kumar
Automated Driving Systems (ADS) rely on AI algorithms, machine learning, and sensor fusion to perform autonomous driving tasks. Safety challenges arise due to the probabilistic behavior of AI/ML algorithms and the need to ensure safety within defined Operational Design Domains (ODDs). Traditional standards such as ISO 26262[3] (Functional Safety) and ISO 21448[4] (SOTIF) address hardware and software failures or functional deficiencies but are insufficient for higher-level autonomous systems (SAE Levels 3–5). To close this gap, additional standards such as UL 4600[1] and ISO 5083[2] provide complementary frameworks for ADS safety assurance. UL 4600[1] establishes a claim-based safety case encompassing the vehicle, infrastructure, and processes, emphasizing structured arguments supported by evidence and reasoning. It offers guidance on autonomy functions, V & V, tool qualification, dependability, and safety culture. ISO 5083[2] focuses on design, verification, and validation of ADS
Mudunuri, Venkateswara RajuAlmasri, HossamFan, Hsing-Hua
Reliable environmental perception under adverse and contaminated conditions is a critical requirement for autonomous driving systems. Although LiDAR sensors play a central role in such perception, their performance is significantly degraded by surface contamination caused by environmental factors such as rain, snow, dust, anti-icing materials, and bug splatter impacts. However, most existing public datasets and prior studies rely on simulated or laboratory-generated contamination scenarios, which limit their applicability to real-world autonomous driving. To address this gap, we construct a large-scale real-world dataset collected from approximately 22,000 km of on-road driving across diverse regions of the United States, covering a wide range of naturally occurring environmental contamination conditions. The dataset was acquired using a multimodal sensing platform integrating LiDAR, perception RGB cameras, infrared camera sensors, and external monitoring systems, enabling
Kim, Hunjae
At present, tire failures directly affect road safety, and the number of incidents caused by them is gradually increasing. Examining wheel attachment loosening on time is vital for vehicle safety. Tire-related incidents not only put people in peril but also have a detrimental effect on the economy. Therefore, the goal of this research is to develop a new and effective method for identifying wheel attachment loosening. A novel gear error reduction approach, distinct from traditional methods, combines advanced computing and probabilistic analysis. This paper involves three key components: extracting looseness eigenvalues, calculating ring gear errors, and computing the tire loosen probabilities. Gear errors derived from the Kalman filter and adjusted for speed, eigenvalues were calculated, and a tire loosening probability analysis was performed. Real-car trials across speeds and roads confirm its accuracy and reliability. This technology can improve automotive safety and maintenance
Liu, JianjianZhang, ZhijieWang, ZhenfengMa, GuangtaoShi, MeijuanLiu, JingZhao, BinggenLu, Yukun
Advanced autonomous driving is a critical component in the intelligent development of new-generation electric vehicles. Research on reliable chassis control algorithms ensures the safety and stability of autonomous vehicles during operation. To enhance the control performance of autonomous vehicles and improve the accuracy of trajectory tracking, this paper proposes a data-driven feedforward compensation trajectory tracking control approach. By optimizing the design of the feedforward compensation loop, systematic errors and latency in the vehicle’s steering system are mitigated, thereby enhancing the precision and robustness of the control algorithm. Initially, the paper analyzes the control errors present when the vehicle responds to controller commands. Subsequently, the paper focuses on the steering angle errors in trajectory tracking, identifying and analyzing the most relevant factors. A time-delay neural network (TDNN) based on data-driven principles is designed to model and
Yang, YijinYuan, YinWang, ZhenfengSu, AilinZhang, ZhijieLu, Yukun
The validation of Advanced Driver Assistance Systems (ADAS) and Automated Driving (AD) Systems, especially at higher automation levels such as SAE Level 3 or 4, demands the testing of a vast array of scenario variants far exceeding the scope of standard safety specifications like Euro NCAP (The European New Car Assessment Programme). Autonomous vehicles require thorough real-world testing to ensure automotive safety. However, public road tests are costly and risky. Instead, virtual scenarios - digital twins of real environments - offer a safe, cost-effective testing alternative. Exhaustive simulation across this high-dimensional scenario space, which includes variations in actor behavior, environmental conditions, and event characteristics, is computationally infeasible. We propose a constraint-solving approach to address this challenge that leverages mathematical and geometric techniques to analytically assess the existence and validity of scenario variants prior to simulation. Two
Karve, OmkarSaurav, SaketPurwar, Prabhanshu
Traffic roundabouts, as complex and safety-critical road scenarios, present significant challenges for autonomous vehicles. In particular, predicting and managing dilemma zone (DZ) encounters at roundabout intersections remains a pivotal concern. This paper introduces an AI-driven system that leverages advanced trajectory forecasting to anticipate DZ events, specifically within traffic roundabouts. At the core of our framework is a modular, graph-structured recurrent architecture powered by graph neural networks (GNNs). By modeling agent interactions as a dynamic graph, our approach integrates heterogeneous data sources - including semantic maps - while capturing agent dynamics with high fidelity. This GNN-based forecasting model enables accurate prediction of DZ events and supports safer, data-driven traffic management decisions for both autonomous and human-driven vehicles. We validate our system on a real-world dataset of roundabout intersections, where it achieves high precision
Lu, DuoSatish, ManthanFarhadi, MohammadChakravarthi, BharateshYang, Yezhou
The exponentially growing complexity of engineering systems, such as robotic systems, autonomous vehicles, and unmanned aerial vehicles, require sophisticated control strategies that can efficiently coordinate system operation in various environments. The traditional control design approaches present significant challenges for control engineers to keep up with the increasing complexity and changing requirements. To advance embedded control system design, a paradigm shift from traditional development approaches toward more structured, systematic methodologies that can manage the multi-domain nature of control systems is critically needed. Model-based design approach is emerging as a solution for this demand. Model-based design approach uses a system model for control system development, from requirements capture to control system design, implementation, and testing. It provides an integrated environment for design, implementation, automatic code generation, and validation, which allows
Repaka, SindhuraChen, Bo
Safety isn’t just the absence of accidents - it’s the presence of trust, empowerment, and accountability at every level. The result is a high-trust culture where process becomes practice and safety is a shared achievement. When people closest to the work feel supported to act on what they see, safety becomes the standard. Thus, the deployment of autonomous driving systems (ADSs) requires not only technical rigor but also a resilient organizational safety culture that supports continuous learning, accountability, and transparent communication. This paper examines how safety culture can be operationalized in ADS development and operations by integrating guidance from standards such as UL 4600 and best practices from SAE AVSC. UL 4600’s requirements for systematic hazard analysis, safety case maintenance, and safety performance indicators (SPIs) are used as a foundation for quantifying organizational behavior within a Just Culture framework. This work draws on Human and Organizational
Wagner, MichaelGittleman, Michele
This study presents the development and validation of a muddy water spray apparatus designed to simulate dust contamination on vehicle sensors for sensor cleaning system testing. It is important to have a constant and quantifiable test environment for the vehicle development process. For verifying the apparatus, muddy water, prepared by mixing standardized dust powder, salt, and water to maintain constant contamination test conditions, was sprayed onto glass specimens to evaluate equipment consistency. Deposited dust weight and thickness were measured across multiple spray cycles, with statistical analyses confirming consistent and reliable deposition. Paired t-tests indicated no significant difference between sample positions, demonstrating uniform spray distribution. The apparatus was further applied to individual infrared (IR) cameras to observe performance degradation under dry and wet contamination conditions showing statistically consistent increases in contamination levels
Jinhyeok, Gong
Autonomous vehicle navigation requires accurate prediction of driving path curvature to ensure smooth and safe trajectory planning. This paper presents a novel approach to curvature prediction using deep neural networks trained on GPS-derived ground truth data, rather than model predictions, providing a more accurate training signal that reflects actual vehicle motion. We develop a multi-modal neural network architecture with temporal GRU encoders that processes vision features, driver intent signals, historical curvature, and vehicle state parameters to predict curvature. A key innovation is the use of GPS-based actual curvature measurements computed from vehicle motion data (κ = ωz/v) as training supervision, enabling the model to learn from real-world driving patterns. The model is trained on 5,322 samples from real-world driving data collected on The University of Oklahoma’s Norman Campus using a Comma 3X device and a 2025 Nissan Leaf electric vehicle. Experimental results
Hajnorouzali, YasamanWang, HanchenLi, TaozheBurch, CollinLee, VictoriaTan, LinArjmandzadeh, ZibaXu, Bin
Accurate perception of the surrounding environment is fundamental and essential to safe and reliable autonomous driving. This work presents an integrated vision-based framework that com bines object detection, 3D spatial localization, and lane segmentation to construct a unified bird’s-eye-view (BEV) representation of the driving scene. The pipeline provides geometric information on object position and orientation by employing Omni3D to infer 3D bounding boxes of objects from monocular camera frames. Detections are subsequently projected onto a 2D BEV canvas, where object instances are represented with respect to the ground plane for enhanced interpretability. To complement the object-level perception, we utilized YOLOPv2 to perform lane segmentation, producing both lane masks and lane line masks in the image domain for future coordinate transformation. By adopting a pinhole camera model, the coordinate transformation of these masks from the perspective image plane into the BEV canvas
Tan, LinArjmanzdadeh, ZibaWang, HanchenLi, TaozheHajnorouzali, YasamanBurch, CollinLee, VictoriaXu, Bin
Electrifying shared autonomous fleets (Robotaxis) presents challenges in balancing decarbonization, service quality, and operational costs, given the limited driving range, long charging times, and suboptimal planning of charging infrastructure. This study develops an integrated energy management and fleet dispatching simulation framework to support cost-effective, low-carbon Robotaxi deployment. The proposed system models both battery electric vehicles (BEV) and internal combustion engine vehicles (ICEV) technologies, and is extensible to other powertrain types. The study also integrates a life cycle assessment module to evaluate well-to-wheel carbon emissions. A total of 1,440 scenarios are designed to test the performance of two service modes (ride-hailing vs. ride-pooling) in terms of energy consumption, emissions, service quality, and operational costs, across varying levels of trip demand and market penetration of different powertrain technologies. The testing aims to verify the
Tang, KangAbdulsattar, HarithYang, HaoWang, Jinghui
Accurately predicting the future trajectories of surrounding vehicles is one of the core tasks in autonomous driving, and its precision is directly related to the safety and reliability of decision-making, path planning, and control execution. However, challenges such as the complexity of traffic participants’ behaviors, the variability of interactions, and the highly dynamic nature of traffic environments make it difficult for existing methods to effectively model spatiotemporal dependencies and achieve accurate long-term prediction in dynamic scenarios, thus limiting their applicability in real-world settings. In this paper, we propose a Transformer-based trajectory prediction model with a spatiotemporal attention mechanism to extract and effectively model vehicle motion and spatial interactions. Specifically, the temporal attention module captures the motion patterns of the target vehicle across the time dimension, while the spatial attention module constructs vehicle interactions
Zhang, LijunHu, XingyuMeng, DejianZhu, Zhehui
This study investigates factors contributing to autonomous vehicle (AV) accidents and proposes an automated fault determination framework. A total of 563 accident reports from the State of California Department of Motor Vehicles spanning from 2019 to 2024 were analyzed by converting unstructured standardized reports into structured data using custom extraction tools. Analysis of these reports reveals that AVs were not at fault in 69.4% of cases and were fully at fault for 22.6% of cases. The proposed method uses these reports to provide an early indicator of fault likelihood and potentially replaces tedious manual review. Machine Learning (ML) and Natural Language Processing techniques were used to replicate the reported faults, achieving 96% average accuracy across three models: Gradient Boosting, Linear Regression, and Random Forest. Through feature engineering techniques in semantic feature extraction from narrative accident descriptions, quantifiable variables were obtained and
Rwejuna, Florida PerfectMajid, NishatulGoutham, MithunLoukili, Alae
The rapid evolution of autonomous vehicle (AV) systems requires scalable, adaptable, and intelligent software architectures to cater for high demands in security, reliability, and real-time processing. This paper introduces a novel software-defined architecture combining generative artificial intelligence (AI) with cloud computing for extending the performance and capabilities of AVs. The proposed methodology uses generative AI models for dynamic perception, route planning, and anomaly detection and is implemented on cloud computing infrastructure to lend orders of magnitude larger computational resources for scaling on-the-fly learning among distributed AV fleets. Decoupling hardware-specific features and transitioning toward a software-defined paradigm, the processing platform allows for quick updates, continuous learning, and flexible deployment of world-leading AI models. Experimental results and simulated scenarios show better situational awareness, response time, and system
Namburi, Venkata Lakshmi
Items per page:
1 – 50 of 3078