Research on Robot Intelligent Control Method Based on Deep Reinforcement Learning

In order to improve the local obstacle avoidance ability of intelligent robots, an obstacle avoidance strategy based on deep reinforcement learning is proposed. The double-stream Q network structure is utilized to train and process the laser ranging data, and the motion data of moving obstacles is used as the observation input, which solves the problem of robot local obstacle avoidance effect in complex, dynamic and unknown environment. Simulation results show that after 1000 simulated obstacle avoidance scenarios, the average reward value of double-stream Q network structure is 780.5, and its average number of moving steps is 809.4. Compared with the other two basic deep reinforcement networks, its learning speed is faster, reward value is higher, and effect difference is more obvious, which indicates that it is more superior in solving the robot local obstacle avoidance problem.