Autonomous Motion Control for Exploration Robots on Rough Terrain Using Enhanced DDPG Deep Reinforcement Learning

The deployment of exploration robots for search and rescue, rather than relying exclusively on human-led efforts, mitigates the risk of property damage and casualties arising from secondary disasters in complex environments. The effectiveness of these robots is often affected by various environmental factors, including the degree of terrain flatness and the presence of obstacles. To address these challenges, we propose a novel approach for autonomous motion control using a deep deterministic policy gradient-based experience (DDPG-E) replay method, which allows exploration robots to navigate autonomously and safely in complex environments. Using deep reinforcement learning to establish the control module, the proposed system enables the exploration robot to generate optimal motion control from environmental information by considering a six-degrees-of-freedom (6-DoF) pose. Experiments indicate that our approach not only promotes navigation with a high level of collision avoidance in complex environments but also achieves higher accuracy and superior generalization ability compared to preceding methods.