Parallel, angular and perpendicular parking for self-driving cars using deep reinforcement learning

Abstract

The progress in creating a fully autonomous selfdriving car has steadily increased in recent decades. Consequently, autonomous parking has been a well-researched field since every driving trip must end with a parking manoeuvre. In recent years, with the current successes in reinforcement learning, the concept of applying it to solve the autonomous parking problem has been more and more explored. A vehicle equipped with a complete autonomous parking system must perform three types of parking: perpendicular, angular and parallel parking. Autonomous parking systems control the steering angle and the vehicle speed by considering the surrounding space conditions to ensure collision-free motion within the available space. This paper presents an approach to the problem of autonomous parking using Reinforcement Learning, more precisely, Deep Deterministic Policy Gradient. This approach proved to be capable of parking in a variety of different environments for the three parking manoeuvres.