Open research questions in Autonomous Vehicle Technology and Safety

What the literature leaves open

• The existing models for vehicle trajectory prediction have limitations in handling non-stationary motion and complex vehicle interactions. The proposed model addresses this gap by combining de-stationary temporal feature enhancement and ego-centric spatial cross-attention.

  A De-Stationary and Cross-Attention LSTM Model for Highway Vehicle Trajectory Prediction · 2026 · DOI

• The limited perception scope of current autonomous vehicles. The need for cooperative autonomous driving and networked sensing and communication. The lack of effective trajectory prediction frameworks for autonomous vehicles.

  IoT-Enabled Cooperative Autonomous Driving: A Hierarchical Spatial–Temporal Transformer Framework for Trajectory Prediction · 2026 · DOI

• Future research can focus on improving the robustness of the method. Future research can explore the application of the method to various traffic scenarios.

  Map-free vehicle trajectory prediction method based on heterogeneous graphs and dynamic scene constraints · 2026 · DOI

• Prior methods have limitations, such as information redundancy and high computational complexity. There is a need for a map-free vehicle trajectory prediction method that can capture dynamic relationships between the target vehicle and surrounding agents.

  Map-free vehicle trajectory prediction method based on heterogeneous graphs and dynamic scene constraints · 2026 · DOI

• Prior work relies on exteroceptive sensors or semantic-based approaches, which have limitations. The complexity of off-road environments and ego-motion uncertainty pose challenges for bumpiness prediction.

  Terrain-Inspired Bumpiness Prediction for Off-Road Autonomous Driving · 2026 · DOI

• There is a lack of robust autonomous driving systems for adverse weather conditions. Conventional sensor fusion methods have limited performance in such conditions.

  E2ETrADS: end-to-end transformer based autonomous driving system for adverse weather conditions · 2026 · DOI

• The gap between the need for safety-critical control laws and the limitations of existing formulations.

  Sensor-Health- and Belief-Aware Risk-Adaptive High-Order Control Barrier Function Safety Filtering for Dynamic Obstacle Avoidance · 2026 · DOI

• Future research can focus on improving the performance of the model on real-world datasets. Further studies can explore the application of the method in different scenarios, such as crowded areas or dynamic environments.

  On Improving Multimodal Pedestrian Trajectory Prediction with CVAE: A Study on Benchmark and Robot Data · 2026

• The task of pedestrian trajectory prediction remains challenging due to the inherent variability and unpredictability of human behavior. Prior methods have limitations in handling uncertainty and adapting to dynamic scenarios.

  On Improving Multimodal Pedestrian Trajectory Prediction with CVAE: A Study on Benchmark and Robot Data · 2026

• The lack of understanding of how drivers behave in naturalistic interactions with pedestrians. The difficulty in designing AV behaviours that are safe, efficient, and accepted by pedestrians.

  Motions Speak Louder Than Words: Designing Human-like AV Speed Profiles at Crosswalks with a Bayesian Optimisation–Driven Adaptive Experiment · 2026 · DOI

• Commonly used rule-based driving metrics like EPDMS are interpretable but lack context-awareness, while recent VLMbased evaluations are context-aware but limited by ambiguous VLM outputs and weak physical grounding.

  DriveJudge: Rethinking Autonomous Driving Evaluation with Vision-Language Models · 2026

• 2-s mean window duration under low-confidence conditions may introduce decision latency problematic in critical takeover situations requiring rapid state transitions; future work should investigate hybrid strategies that impose upper bounds on window duration when takeover-critical contextual signals are detected.

  Confidence-driven adaptive time window for real-time driver fatigue detection in Level 2-3 autonomous vehicles: a multi-dataset validation study · 2026 · DOI

• Prevailing data-driven models lack human-like, interpretable reasoning. The gap is that prevailing models neglect the WHY behind actions. The gap is that prevailing models do not integrate cognitive science principles.

  Embodied cognition-driven interpretable trajectory prediction of autonomous systems · 2026 · DOI

• Pure E2E models lack a viable regulatory path for 2026 L4 deployment due to prohibitive validation mandates. The study does not provide a detailed analysis of the economic and social implications of autonomous vehicles. The paper does not discuss the potential risks and challenges associated with the development and deployment of autonomous vehicles.

  Driving into the Future: A Comprehensive Survey of Autonomous Vehicle Technologies · 2026 · DOI

• The lack of a comprehensive survey of autonomous vehicle technology. The need for a viable solution for autonomous vehicle development that addresses the limitations of pure E2E models. The gap in the existing literature on the evaluation of autonomous vehicle software architectures.

  Driving into the Future: A Comprehensive Survey of Autonomous Vehicle Technologies · 2026 · DOI

• Although various ADS testing techniques have been proposed to identify accident-prone driving scenarios, reliable ADS-oriented accident diagnosis techniques remain scarce.

  ADSDx : Towards Automated Accident Diagnosis for High-level Autonomous Driving Systems · 2026 · DOI

• Need for a comprehensive dataset for instruction-driven traffic video segmentation in adverse winter environments.

  SnowRef-Drive: A Global Instruction-Driven Traffic Video Segmentation Dataset for Adverse Weather Driving Scenarios · 2026 · DOI

• Existing two-wheeler safety systems focus on static compliance checks (helmet detection, license plate recognition) but do not jointly model dynamic riding events and behavioral patterns in real time. No framework integrates event-based riding behavior recognition (acceleration, braking, maneuvers) with safety compliance monitoring under unstructured traffic conditions.

  Robust Multi-Task Framework for Motorcycle Rider Safety in Unstructured Traffic · 2026 · DOI

• Several directions for future enhancement are identified: 1. Accident Detection: Incorporating LSTM or transformer-based temporal sequence models to detect abnormal vehicle behavior patterns—including sudden stops, abrupt lane changes, and collision trajectories—to enable proactive accident detection and automated emergency dispatch.

  IoT-Enabled Speed and Accident Detection Platform Using Deep Learning and Multi-Object Tracking · 2026 · DOI

• This technical volume demonstrates that through precise sensor stabilization and adaptive resolution mapping, a robust DMS can operate effectively within a resource-constrained Edge-AI environment. Enforcing fixed hardware gains completely bypasses the dynamic visual noise that typically degrades deep learning models in automotive applications.

  Technical Report: Edge-AI Optimization and Sensor Stabilization for Automotive Driver Monitoring Systems (DMS) · 2026 · DOI

• Limited available reaction time in dynamic obstacle avoidance scenarios.

  Risk aware safe control with multi-modal sensing for dynamic obstacle avoidance · 2026 · DOI

• Balancing accuracy with low-latency inference. Deploying ADAS solutions on edge hardware. Addressing the gaps in existing ADAS solutions.

  Edge deployable lightweight framework for real-time driver safety system with integrated lane departure and traffic sign recognition · 2026 · DOI

• Existing reinforcement learning methods fail to fully account for heterogeneous risk exposure and non-stationary, multi-agent interactions. Many RL models assume stable dynamics and risk-neutral optimization, which rarely hold in open-world driving.

  Risk-Aware Reinforcement Learning for Cooperative Autonomous Vehicle Coordination with Adaptive Risk Sensitivity and Multi-Agent Optimization · 2026 · DOI

• Further evaluation of the proposed framework in real-world scenarios. Investigation of the applicability of the minimum-jerk model to other types of road interactions.

  Motions Speak Louder Than Words: Designing Human-like AV Speed Profiles at Crosswalks with a Bayesian Optimisation–Driven Adaptive Experiment · 2026 · DOI

• Developing more robust and accurate traffic video segmentation models using SnowRef-Drive. Evaluating the performance of different models and algorithms in adverse winter environments. Exploring the application of SnowRef-Drive to other adverse weather conditions.

  SnowRef-Drive: A Global Instruction-Driven Traffic Video Segmentation Dataset for Adverse Weather Driving Scenarios · 2026 · DOI

Most-cited papers in Autonomous Vehicle Technology and Safety

• A Survey of Deep Learning Applications to Autonomous Vehicle Control · IEEE Transactions on Intelligent Transportation Systems · 2021 · 710 citations
• Perception and sensing for autonomous vehicles under adverse weather conditions: A survey · ISPRS Journal of Photogrammetry and Remote Sensing · 2023 · 587 citations
• Motion Planning for Autonomous Driving: The State of the Art and Future Perspectives · IEEE Transactions on Intelligent Vehicles · 2023 · 577 citations
• Survey of Deep Reinforcement Learning for Motion Planning of Autonomous Vehicles · IEEE Transactions on Intelligent Transportation Systems · 2022 · 575 citations
• End-to-End Autonomous Driving: Challenges and Frontiers · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2024 · 457 citations
• Modeling Driver Behavior in a Cognitive Architecture · Human Factors The Journal of the Human Factors and Ergonomics Society · 2006 · 416 citations
• Research Advances and Challenges of Autonomous and Connected Ground Vehicles · IEEE Transactions on Intelligent Transportation Systems · 2021 · 353 citations
• Intelligent driving intelligence test for autonomous vehicles with naturalistic and adversarial environment · Nature Communications · 2021 · 352 citations
• A feature learning approach based on XGBoost for driving assessment and risk prediction · Accident Analysis & Prevention · 2019 · 336 citations
• Multi-Class Road User Detection With 3+1D Radar in the View-of-Delft Dataset · IEEE Robotics and Automation Letters · 2022 · 308 citations

Most recent work

• GIF-Calib: Geometry-intensity fusion for 4D radar-camera calibration in autonomous driving vehicles · Information Fusion · 2026
• Automating the estimation of turning movement rates at multilane roundabouts using unmanned aerial vehicles and deep learning · Green Energy and Intelligent Transportation · 2026
• BANet: Bidirectional Feature Aggregation and Adaptive Multi-Scene Perception-Based Lane Detection for Autonomous Driving · IEEE Transactions on Intelligent Transportation Systems · 2026
• PlanAgent: A Multimodal Large Language Agent for Closed-Loop Vehicle Motion Planning · IEEE Transactions on Cognitive and Developmental Systems · 2026
• A Survey on the Application of Large Language Models in Scenario-Based Testing of Automated Driving Systems · IEEE Transactions on Intelligent Transportation Systems · 2026
• Criticality Assessment Model for Intelligent Vehicle Test Scenario Based on Interactive Field Feature and Hypergraph Learning · IEEE Transactions on Intelligent Transportation Systems · 2026
• SDATNet: Social Diverse Adaptive TrajectorNet for Pedestrian Trajectory Prediction Under Long-Tail Behavior Distributions · IEEE Transactions on Intelligent Transportation Systems · 2026
• Generative AI for Testing of Autonomous Driving Systems: A Survey · ACM Transactions on Software Engineering and Methodology · 2026
• A hierarchical trajectory planning framework based on diffusion models for autonomous vehicles · Advanced Engineering Informatics · 2026
• DRO-EDL-MPC: Evidential Deep Learning-Based Distributionally Robust Model Predictive Control for Safe Autonomous Driving · IEEE Robotics and Automation Letters · 2026

Related topics in Engineering