Towards Accountable Self-Driving Cars: A Framework for Identifying the Cause of Autonomous Traffic Collisions

Wide adoption of Autonomous Vehicles (AVs) will ultimately stem from their ability to make timely, reliable, and safe decisions, resulting in fewer collisions and overall harm. However, in their current state, AVs are still capable of misbehaving and causing car accidents. In this paper, we develop a framework for AV accident cause identification. Our framework makes use of the Fast Causal Inference (FCI) algorithm to extract an explanation for the accident given readings, velocity and steering data from the vehicles involved. The proposed framework promotes trust in AVs and supports legal personnel and developers to analyze accidents, creating a fertile ground for better and safer AVs. The framework is evaluated using computer simulations of realistic self-driving car accidents from the California Department of Motor Vehicles (DMV).