Mobile Robot Navigation in an Unknown Environment Based on Deep Reinforcement Learning

This paper studies an end-to-end navigation method based on deep reinforcement learning, which enables mobile robots to independently explore and navigate in unknown environments using only sensor data and without map information. In this paper, we propose an improved deep reinforcement learning method, design a deep reinforcement learning (DRL) optimization model, optimize the experience pool to improve the utilization rate of valuable positive reward transfer samples, and introduce a goal orientation strategy to help the robot avoid blind exploration. Experimental results show that the proposed optimization model could realize path planning and dynamic obstacle avoidance for mobile robots in complex environments. The convergence speed, planning success rate, and path precision were improved to a certain extent. Compared with the traditional DRL method, the robot's exploration efficiency was higher, and the learned strategies were more reliable.