Reinforcement Learning pro autonomní parkování

Abstract

This master thesis explores the application of reinforcement learning methods to autonomous parking tasks. The thesis focuses on the implementation of a parking simulation environment for testing autonomous agents using the Unity game engine and the ML-Agents library. The thesis first provides an overview of reinforcement learning methods, including deep reinforcement learning algorithm Proximal Policy Optimization and imitation learning algorithm Generative Adversarial Imitation Learning. The current state-of-the-art technologies and approaches in autonomous parking are also discussed, followed by the implementation of the simulation environment and the specification of the rewards and training process of autonomous agents. The experimental results demonstrate the effectiveness of reinforcement learning based approaches to autonomous parking tasks in various scenarios, including fixed and random targets and parallel parking.