Advancements in Autonomous Driving Safety and Scenario Generation

The field of autonomous driving is rapidly advancing, with a focus on improving safety and robustness. Recent research has highlighted the importance of generating realistic and diverse scenarios for testing and validation, as well as the need for more comprehensive safety analysis frameworks. One notable trend is the integration of large language models and generative AI techniques to create more realistic and controllable scenarios, such as LD-Scene, which uses latent diffusion models and large language models to generate adversarial safety-critical driving scenarios. Another area of focus is the development of safety-critical scenario libraries, such as Safety2Drive, which provides a comprehensive test framework for autonomous driving systems. Additionally, researchers are exploring new approaches to safety analysis, including the use of dynamic Bayesian networks to model human cognitive states and trust in mobility, as seen in the Toward Informed AV Decision-Making paper. Noteworthy papers in this area include LD-Scene, which achieves state-of-the-art performance in generating realistic and diverse adversarial scenarios, and Safety2Drive, which provides a paradigm-shifting test framework for safe deployment of autonomous driving systems.

Sources

LD-Scene: LLM-Guided Diffusion for Controllable Generation of Adversarial Safety-Critical Driving Scenarios

Safety2Drive: Safety-Critical Scenario Benchmark for the Evaluation of Autonomous Driving

D4+: Emergent Adversarial Driving Maneuvers with Approximate Functional Optimization

On-Demand Scenario Generation for Testing Automated Driving Systems

Multi-modal Traffic Scenario Generation for Autonomous Driving System Testing

Toward Informed AV Decision-Making: Computational Model of Well-being and Trust in Mobility

UniSTPA: A Safety Analysis Framework for End-to-End Autonomous Driving

Generative AI for Autonomous Driving: A Review

Behavioral Safety Assessment towards Large-scale Deployment of Autonomous Vehicles

A Survey on the Application of Large Language Models in Scenario-Based Testing of Automated Driving Systems

Built with on top of