Research On Robot Trajectory Planning Method Based On Deep Reinforcement Learning

At present,with the development of deep learning and other technologies,the intelligent level of robot systems is getting higher and higher,and good applications have been achieved in industry,medical and other industries.People expect robots to enter people’s lives,coexist with people,and provide services for people in the future.This requires robots not only to have good mobility and operation capabilities,but also to have the ability to adapt to complex unstructured environments such as human settlements.It can automatically identify the system environment,and dynamically adjust its movement and operation trajectory according to environmental changes.Trajectory planning in an unstructured environment is one of the difficulties in robotics research.Based on the comparative analysis of existing robot trajectory planning methods,this paper introduces deep reinforcement learning into robot trajectory planning to carry out research on robot trajectory planning in unstructured dynamic environments.On the one hand,for the two-dimensional plane motion of the mobile robot,the artificial potential field method and the improved double depth Q network are combined to realize the trajectory planning of the unstructured dynamic environment of the mobile robot;on the other hand,for the three-dimensional space of the six-degree-of-freedom robotic arm Motion,proposes a robotic arm trajectory planning method based on soft actor-critic and g SDE.On this basis,this paper further investigates the problems of inefficiency and sparse algorithmic rewards of discrete and continuous deep reinforcement learning in unstructured dynamic scenarios.The main work of this paper is as follows:Firstly,this paper introduces the research status of robot dynamic trajectory planning and deep reinforcement learning at home and abroad,expounds the application of deep reinforcement learning method in robot dynamic trajectory planning,and analyzes and compares the advantages and disadvantages of classical planning theory and deep reinforcement learning planning method.Then,the kinematic model of the robot is analyzed,and the artificial potential field method,deep reinforcement learning method and g SDE method of robot dynamic trajectory planning are compared and studied.Secondly,on this basis,combining the deep reinforcement learning method with the traditional trajectory planning method,the PF-IDDQN four-wheeled mobile robot two-dimensional plane motion trajectory planning method is proposed;combining the efficient exploration method with the deep reinforcement learning method,the g SDE-SAC’s six-degree-of-freedom manipulator arm three-dimensional space motion trajectory planning method.The above two methods improve the reward function value of the algorithm,the success rate of trajectory planning and the exploration efficiency of the robot,so that the robot can move and operate autonomously in an unstructured dynamic environment.Finally,the two improved algorithms proposed in this paper are verified on a simulation platform based on a four-wheeled mobile robot and a six-degree-of-freedom manipulator,and a simulation experimental environment and experimental tasks are designed to evaluate the performance of the improved algorithms.The results show that the PF-IDDQN algorithm proposed in this paper enables the mobile robot to reach the target position within a limited number of explorations in an unstructured dynamic environment.The experimental success rate of the algorithm is over 97%,and the average reward value is increased by 230.The task cannot be completed;based on the g SDE-SAC algorithm proposed in this paper,the robotic arm can grasp the target object in an unstructured dynamic environment and plan a collision-free trajectory to the target position.Compared with the classical SAC algorithm,the algorithm proposed in this paper has The experimental success rate and reward function value are increased by about 12% and 19%,respectively,verifying the effectiveness of the algorithm.