The Problem of Many Vehicles: An Explainable System for Autonomous Multi-agent Accidents

Given the need for trustworthy autonomous vehicles, this research paper presents an explanatory framework for multi-agent autonomous vehicle crashes, leveraging the FCI causal algorithm to generate a Full Time Causal Graph (FTCG). The framework’s performance is evaluated using simulated scenarios within the MetaDrive environment, employing key metrics such as accuracy, time, scalability, and stability. Results demonstrate promising accuracy in crash explanation, showcasing the effectiveness of the FCI causal algorithm. The scalability analysis reveals that the framework operates within reasonable time constraints as the dataset size increases. However, the introduction of uniform noise poses a challenge to stability, adversely affecting the system’s reliability in identifying crash sequences under random variations. While the framework exhibits strengths in accuracy and scalability, addressing stability concerns, especially in the presence of uniform noise, presents an avenue for future research and improvement.