Responsibility-Sensitive Collision Risk Assessment and Maneuvering Safety Evaluation

Safety is one important basis for autonomous vehicles being accepted by the human society. Ideally, autonomous vehicles are expected to never cause accidents. However, the safety is somehow antagonistic to efficiency. Considering the real-world complex traffic conditions, e.g., traffic jams, drivers ignoring traffic rules, etc., a zero-accident transportation system is barely possible. To balance between traffic safety and efficiency, autonomous vehicles should be able to estimate the scenario risk and adjust its maneuvering strategy accordingly. This paper proposes a method to quantitively assess the collision risk of two vehicles during interactions. In other words, it measures the chance of a vehicle responsible for an accident. The risk is decided by the complexity of the interaction scenario, and thus is termed as responsibility-sensitive collision risk of the scenario. Based on the scenario risk assessment, a safety evaluation method is proposed to measure to what degree the maneuvering strategy is safe. It not only assesses what risks the target vehicle can afford, but also measures the relative safety capability compared to other vehicles in a statistical sense. The risk and safety evaluations also help to understand the maneuvering styles, i.e., aggressive or conservative.