A Deep Reinforcement Learning-based Robot Navigation Method using Sensor Information Fusion

This paper proposes a Deep Reinforcement Learning (DRL)-based method using multi-sensor information fusion for mobile robot navigation tasks. In order to leverage the precise distance data of the 2D LiDAR and the rich information of the camera image, the proposed method improves the ability of the LiDAR to identify obstacles by leveraging the obstacle information in the camera image, while ensuring that there is no coupling between the image and the LiDAR data. Furthermore, a self-attention network is introduced to enhance the perception of the current environment. The proposed method is evaluated through simulations and a navigation framework based on DRL, which is designed to test and apply perception effects. The experimental results demonstrate that the proposed method enhances the perception and expression ability of the laser radar to obstacles, thereby effectively improving the performance of the robot to avoid obstacles.