Research And Application Of Agents Obstacle Avoidance And Path Planning Based On Deep Reinforcement Learning

With the vigorous development of scientific theory and technology,and the increasing demand for science and technology in humans' lives,the technology of obstacle avoidance and path planning of agents has attracted wide attention.It has a great significance in the practice of life production.However,early researches focused more on the path planning of single agent in a static environment,and the path cooperative planning of multi-agent in dynamic environment was neglected.With the complexity of application scenarios,multi-agent obstacle avoidance and path planning have played an increasingly important role in life's applications.Just considering some simple situations,it is no longer suitable for the development requirements now.At the same time,how to make the multi-agent system more intelligent,more self-learning ability,more adaptable to different change scenarios,is also the hotspot and future development trend of recently researches.In order to achieve the above goals,based on the previous research results and combines with the method of deep reinforcement learning,this thesis studies obstacle avoidance and path planning of multi-agent in many aspects,and the simulation experiment is carried out and some results are obtained.It can be summarized as follows:(1)This thesis proposed a muti-agent reciprocal obstacle avoidance algorithm based on deep reinforcement learning,which considering the priority of each agent in obstacle avoidance.The method is based on MADDPG multi-agent deep reinforcement learning framework,and the state spaces,action spaces and a reward function combined with ORCA method for obstacle avoidance are set up.Through a standard reinforcement learning process,the reciprocal obstacle avoidance between two agents is completed in the simulation experiment.At the same time,the time index is set and compared with ORCA method,which shows the efficiency of this method.(2)The algorithm of obstacle avoidance between two agents is extended to multiagent obstacle avoidance and path planning.At the same time,the obstacle is regarded as an agent with infinite priority,and the path planning problem with complex obstacle environment is simplified to a multi-agent obstacle avoidance problem,and the effectiveness of the method is verified,and the advantages of this method are illustrated by comparing the method with ORCA method.(3)The deep reinforcement learning method is applied to the multi-task multi-agent path planning problem.The task assignment process and the path planning process are combined to complete the multi-agent multi-task path planning.A simulation scenario is selected and the simulation experiment is carried out to verify the effectiveness of the method.In summary,this thesis uses the method of deep reinforcement learning to study and deal with multi-agent obstacle avoidance and path planning problems,and achieves good results through simulation experiments.