Acquisition of pareto optimal policies for collision avoidance in mixed traffic flow

Abstract

Autonomous driving is expected to reduce the number of traffic accidents caused by human error, but it has only been able to handle simple situations such as driving on highways. However, the demand for autonomous driving is for safe and efficient driving in mixed traffic flows. We focus on autonomous driving in situations where multiple moving obstacles exist simultaneously. Model-based control is problematic because it is difficult to construct a driving environment model in these situations. Therefore, we introduce a reinforcement learning method that does not require a driving environment model. This paper proposes the collision avoidance problem as a multi-objective sequential decision-making problem. We propose a method for learning Pareto-optimal driving policies concerning safety and speed using multi-objective reinforcement learning. We verified the performance of the proposed method by computer experiments in a T-intersection environment and confirmed the acquisition of multiple Pareto-optimal driving policies.