Xin Wang∗ and Peter Xiaoping Liu∗∗
Mobile robot, path planning, obstacle avoidance, artificial potential field.
Path planning and obstacle avoidance are critical tasks for mobile robots. In this study, we propose an enhanced approach based on the double dueling deep Q network (D3QN) to address these challenges. This approach combines the artificial potential field method with PER to enable collision-free path planning in complex environments. The presented method aims to achieve optimal decision-making and ensure safe navigation. Specifically, the integration of PER in D3QN mitigates the issue of low learning efficiency during mobile robot training, thereby accelerating convergence toward optimal decisions. Furthermore, to overcome challenges such as sparse rewards and the proximity of mobile robot to obstacles, we introduce an artificial potential field, which improves the reward function in D3QN and accelerates the learning process. Simulation experiments verified the superiority of the presented method over traditional D3QN, demonstrating higher learning efficiency, faster convergence, and the ability to ensure collision-free paths that maintain a safe distance from obstacles.
Important Links:
Go Back
