Research On Path Planning Of Elderly Service Robot Based On Particle Swarm Optimization And Reinforcement Learning

At present,as Chinese aging degree continues to intensify,research on service robots for helping the elderly has received extensive attention from scientific researchers.Path planning technology is an important research direction of service robots for the elderly.Although important results have been achieved in the research on path planning of service robots for the elderly,there are still limitations.The service robot for the elderly is in a dynamic and uncertain working environment.Due to the lack of global prior information in the online path planning method,the path obtained not be optimal;although the offline path planning method can obtain the global optimal path based on the global information.However,the real-time performance is poor and obstacle avoidance is not possible dynamically.The main research content is aimed at the problem of obtaining the optimal path for the elderly service robot's path planning in a dynamic and uncertain environment,using an "offline" and then "online" path planning method.The main research work is as follows:(1)Aiming at the problem of the basic Particle Swarm Optimization(PSO)offline path planning,it is difficult to balance the global exploration and local development capability,easy to fall into local optimum.An improved particle swarm optimization algorithm,Random Disturbance Self-adaptive Particle Swarm Optimization(RDSPSO)is proposed for global optimization.The environment performs offline path planning,introduces a distance evolution factor to adopt a random perturbation strategy for the particles in the search performance degradation area,and uses an adaptive strategy to intervene in the three control parameters of the algorithm to dynamically adjust the global exploration and local development capabilities.Experiments show that the RDSPSO algorithm overcomes the problems of the PSO algorithm and verifies the effectiveness of the algorithm.(2)Aiming at the problem of insufficient storage space and dimension disaster in Q-value table of Q-learning algorithm online path planning in dynamic environment,an improved Q-learning algorithm is proposed.The robot motion state scheme is defined by the target position information and the nearest obstacle information,which simplifies the Q-value table established by the motion state action index.Experiments show that the improved Q-learning algorithm solves the problems of online path planning in dynamic environment and improves the work efficiency.(3)Aiming at the low efficiency of Q-learning algorithm in dynamic environment,a continuous reward function is designed to make every action taken by the service robot get the corresponding reward.In addition,due to the limitation of robot dynamics,the dynamic window of robot dynamics is introduced to calculate the action of each state.Experiments show that the design method is effective and feasible,and improves the efficiency of algorithm training.(4)Aiming at the problem of obtaining the optimal path for the elder service robot in path planning in a dynamic and uncertain environment,first use the RDSPSO algorithm to perform offline path planning on the global environment,plan the initial path of the elder service robot,and help the elder service robot follow the initial path When walking,when the robot sensing device detects that the distance between the moving obstacle and the robot is less than the design threshold,the improved Q-learning algorithm is used to re-plan the online path to avoid dynamic obstacles.After reaching the safe distance range,the elderly service robot returns to the initial path.Continue to run to the end to complete the task.Experiments show that this planning method improves the quality of the planned path and reduces the planning time at the same time.