Towards End-to-End Chase in Urban Autonomous Driving Using Reinforcement Learning

Michal Kolomanski, Mustafa Sakhai, Jakub Nowak, Maciej Wielgosz

Abstract: This paper addresses the challenging task of developing an autonomous chase protocol. First, training of an autonomous vehicle capable of driving autonomously from point A to B was developed to proceed with a chase protocol as a second step. A dedicated driving setup, based on a discrete action space and a single RGB camera, was developed through a series of experiments. A dedicated curriculum learning agenda allowed to train the model capable of performing all fundamental road maneuvers. Several reward functions were proposed, which enabled effective training of the agent. In the subsequent experiments, we selected the reward function and model that produced the most significant outcome, guaranteeing that the chasing car was within 25 m of a runaway car for 63% of the episode duration. To the best of our knowledge, this work is the first one that addressed the task of the chase in urban driving using the Reinforcement Learning approach.