Autonomous Navigation of Unmanned Vehicle Through Deep Reinforcement Learning
Letian Xu, Hao Liu, Haopeng Zhao, Tianyao Zheng, Tongzhou Jiang, Lipeng Liu
Abstract
This paper explores the use of Deep Reinforcement Learning (DRL) to achieve autonomous navigation for unmanned vehicles, with a focus on the Deep Deterministic Policy Gradient (DDPG) algorithm. The main challenge addressed is handling high-dimensional, continuous action spaces, which are commonly encountered in autonomous navigation tasks. The study presents the Ackermann robot model used for testing and provides an explanation of how the DDPG algorithm is applied to the navigation problem. Through experiments conducted in a simulation environment, the feasibility and effectiveness of the proposed approach are verified. The results indicate that the DDPG algorithm outperforms traditional reinforcement learning algorithms, such as Deep Q-Network (DQN) and Double Deep Q-Network (DDQN), particularly in path planning tasks. The improved algorithm demonstrates better decision-making, leading to more accurate and reliable navigation in comparison to the traditional methods.