Research On Robot Motion Control Algorithm Based On Deep Reinforcement Learning

With the continuous development of artificial intelligence and robot technology,the application field of robot is also constantly expanding.Ensuring stable and safe motion control for robots during task execution has always been a key focus in robotics research.Robot motion control is not only the solid foundation for efficient task execution by robots,but also the key to achieving flexible and intelligent robot operation.With the deepening of research,robot motion control based on deep reinforcement learning has achieved remarkable results.In practical applications,deep reinforcement learning has problems such as unstable control strategy in the initial training period,low learning efficiency and long training time.These problems restrict the application of deep reinforcement learning in robot motion control.This paper studies the robot motion control based on deep reinforcement learning algorithm,aiming to improve the stability and adaptability of robot motion control by enhancing the deep reinforcement learning algorithm.The main work of this paper is as follows.Based on the research of deep deterministic policy gradient algorithm for robot motion control,a deep deterministic policy gradient algorithm based on adaptive experience filtering is proposed.In order to address the issues of long training time and low learning efficiency in deep deterministic policy gradient algorithm,the experience replay mechanism is improved.Utilizing an adaptive experience filtering strategy to fully exploit the experience data in the experience buffer.The calculation method of experience data priority is designed,and the experience filter is used to distribute the experience data to different experience buffers.Then,the experience selector is used to extract the experience data from experience buffer for the training of the model,which improves the utilization efficiency of the experience data and enhances the exploration ability of the robot.Compared with mainstream deep reinforcement learning algorithms,the experimental results demonstrate that the deep deterministic policy gradient algorithm based on adaptive experience filtering can effectively accelerate the learning efficiency and training speed of robots in the motion simulation environment,achieving more stable motion control.Aiming at the problems of unstable control strategy and long training time in the early stage of the proximal policy optimization algorithm,a proximal policy optimization algorithm based on attention mechanism structure decomposition is proposed.Since different joints have different influence weights in robot motion control,attention mechanism is introduced into the strategy network to decompose the robot structure.By focusing on the important joint structure of the robot,more efficient and accurate motion control can be achieved.According to the complexity of robot action and state space,the strategy network is designed as multiple subnetworks.Each subnetwork is responsible for processing the decomposed joint state action information.The initialization mode of strategy network and value network is designed.And the adaptive experience filtering strategy is used to replace the traditional experience replay mechanism to improve the stability of robot training.Compared with the baseline algorithm,the experimental results show that the proposed algorithm can enhance the generalization ability of the model to some extent,and improve the stability and convergence speed of the robot motion control.