Research On Dynamic Path Planning Of AGV Based On Reinforcement Learning

Traditional Automated Guided Vehicles(AGV)path planning needs to provide detailed scene maps,which are not suitable for complex dynamic environments,resulting in low work efficiency and prone to multi-vehicle jams.Because reinforcement learning has the characteristics of autonomous learning,dynamic adjustment,etc.,it can complete dynamic path planning tasks and obtain the optimal path without an environmental model.However,this paper aims at the exploration of reinforcement learning algorithms-the use of imbalance and continuous state space,and proposes a reinforcement learning algorithm(K-L)based on Kohonen network fusion.And design a two-layer path planning method(GA-KL)based on the combination of improved genetic algorithm and K-L algorithm,which can achieve rapid algorithm convergence in a dynamic and complex environment and ensure collision-free operation of AGVThe main work is as follows:1)This paper designs an AGV path planning algorithm(K-L algorithm)based on improved reinforcement learning.First,the traditional exploration-utilization balance method is improved,and the simulated annealing algorithm is introduced to optimize the value of ? in ?-greedy to increase the ability to jump out of the local optimum,thereby improving the imbalance between exploration-utilization and preventing frequent AGVs.Detours cannot find the optimal solution,and multiple explorations lead to too long learning time;then,based on the optimization of the simulated annealing algorithm,the continuous external environment is clustered and analyzed by fusing the Kohonen network to obtain a state generalization representation method,to improve the problem of slow convergence speed of traditional reinforcement learning algorithm and low efficiency of AGV environment learning.Finally,simulations and actual experiments verify the effectiveness of the algorithm and its advantages in convergence time.2)Based on the K-L algorithm,this paper designs a two-level path planning method(GA-KL)based on the combination of improved genetic algorithm and K-L algorithm.First,the genetic algorithm(T-GA)optimized by the three-crossover operator and the node busy factor is used as a global planning algorithm.The topology map node is used as an encoding to avoid static obstacles with fixed coordinates to generate a global optimal path node set;The A*algorithm is used to find the shortest path between nodes with the shortest path as the goal.Then,if there are no dynamic obstacles on the running path,the path planned by the A*algorithm is used to pass quickly;if a dynamic obstacle is found,the above-mentioned K-L algorithm is used for local planning between nodes to avoid local dynamic obstacles.This method combines T-GA algorithm,A*algorithm and K-L algorithm,and has good global search and dynamic avoidance capabilities.Simulation experiments and real vehicle experiments show that in a complex dynamic environment,the GA-KL-based path planning method can achieve effective dynamic obstacle avoidance and advantages in convergence time.