Research On Path Planning Of Multi-assisted Robot Based On Deep Reinforcement Learning

In recent years,the aging of our country has become more and more serious and the number of patients with lower extremity disability is increasing year by year.Manual nursing can no longer meet the needs.The use of assistive robots to assist elderly and disabled patients in their daily life is a hot research topic.When there are multiple auxiliary robots in the environment,the path planning and obstacle avoidance of multiple auxiliary robots is a primary problem to be solved.Multi-robot systems have high dynamics,so traditional path planning methods are not suitable for multi-robot systems.How to make path planning in multi-robot systems more intelligent is the focus of current research.In recent years,the popularity of machine learning has become increasingly high.The mechanism of reinforcement learning in machine learning to achieve optimal strategies through continuous trial and error has attracted the attention of many scholars.In this thesis,the deep reinforcement learning method is used to study path planning and multi-robot obstacle avoidance in the multi-assisted robot system.The main contents of this thesis are as follows:First of all,it analyzes the particularity of the users of the assistant robot in the old-age and disabled scene and puts forward three basic requirements of safety,speed,and comfort for the path planning of the assistant robot.The deficiencies of traditional robot path planning methods in the use of assisted living robots are pointed out.Applying Deep Q-Networks in Deep Reinforcement Learning to Assist Robot Path Planning.A simulation experiment of auxiliary robot path planning in the scenario of the intelligent nursing home is designed.Compared with the VFH path planning method,it is found that the path planning method based on deep reinforcement learning shortens the path planning time,improves the smoothness of the trajectory,and better meets the requirements of safety,rapidity,and comfort of the auxiliary robot path planning.Then,in the process of using deep reinforcement learning,it is found that its training efficiency is slow and the iterative process requires many time steps.The reasons for such deficiencies are analyzed,and a resampling mechanism is proposed to improve the playback probability of samples with important information value.Apply resampling mechanism to deep deep reinforcement learning.The simulation experiment of auxiliary robot path planning is carried out in the scenario of the intelligent nursing home.Compared with the traditional deep Q network,it is proved that the resampling mechanism can shorten the iterative steps of deep reinforcement learning,reduce the time steps in the iterative process and improve the training efficiency of deep reinforcement learning.Finally,a multi-aided robot system is established,the obstacle avoidance requirements of the multi-aided robot system are analyzed,and the mathematical model of multi-robot obstacle avoidance is established.In view of the low algorithm efficiency of the classical multi-robot obstacle avoidance method and the poor adaptability to large-scale robot systems.A multi-robot obstacle avoidance method based on multi-agent deep reinforcement learning is designed.In the multi-robot obstacle avoidance experiment in extreme cases,comparing the results of the speed obstacle method,the optimal mutual obstacle avoidance method,and the multi-agent deep reinforcement learning method,it is found that the robot obstacle avoidance displacement of the multi-agent deep reinforcement learning method is shorter.,the speed of the robot changes more smoothly,and the application of the multi-agent deep reinforcement learning method to the largescale robot system can still maintain stable obstacle avoidance.The research shows that the multirobot obstacle avoidance method based on deep reinforcement learning is a feasible and highquality obstacle avoidance method for multi-aided robots.