Enhancing Autonomous Navigation: Integration of Artificial Potential Field-Based Reward Functions and Priority Experience Replay

In recent years, significant advancements in the field of autonomous navigation for unmanned vehicles have been achieved by applying new deep reinforcement learning methods. However, deep reinforcement learning still faces numerous challenges in autonomous navigation, such as sparse rewards and insufficient utilization of experience. To address above issues, this paper introduces a new algorithm named Artificial Potential field-Enhanced Reinforcement Navigation (APERN). APERN incorporates a reward function design based on the artificial potential field method. This design guides unmanned vehicles to effectively avoid obstacles and reach the target point, addressing the issue of sparse rewards during the exploration process. Concurrently, a priority experience replay mechanism is used in the learning process of the proposed algorithm to sample experiences, which enhances the exploitation of scarce experiences while adjusting priorities to prevent overfitting. Experiments conducted in the Gazebo simulation environment using the APERN algorithm, show that APERN algorithm increases the navigation success rate by approximately 14% over the popular navigation method Goal-Driven Autonomous Exploration (GDAE). Moreover, the model converges more rapidly and exhibits stronger adaptability in complex environments.