Research on Mobile Robot Path Planning Based on Deep Reinforcement Learning

摘要

Path planning based on deep reinforcement learning has been a hot topic in the field of mobile robots in recent years, but there are still many shortcomings in its application. Such as the lack of agents' generalization ability, the loss of targets in the process of exploration and the inefficiency of data exploration. In this paper, we propose an improved path planning algorithm to solve the above three problems. We take the Depth Deterministic Policy Gradient (DDPG) algorithm as the basic algorithm, and increase generalization ability of agents by adding adaptive gaussian noise, then, in order to solve the problem of losing target in exploration process, we redesign the reward functions based on the curiosity mechanism. We also add the memory units for agents to make the exploration process more efficient. Finally, we verify that the improved algorithm has good adaptability in different obstacle avoidance environments through simulation experiments.