Research On Reinforcement Learning Algorithm For AGV Path Plannin

Automatic guided vehicle(AGV)is a kind of handling equipment for management automation of storage.With the development of automation in logistics industry,the demand for AGV is increasing.AGV is one of the mobile robots,which can replace manpower for intelligent cargo handling,improve operation efficiency,and have the advantages of safety,reliability and flexibility.Therefore,it has been widely used.This paper mainly studies the improved A star algorithm for AGV global path planning,the enhanced ant colony algorithm based on reinforcement learning for local path planning,and the improved Q-learning algorithm for path planning in unknown environment.The main work of this paper includes the following aspects:1.The modeling methods of map environment such as topological map method,visual map method,grid map method and free space method are studied.The grid map method is determined after discussing their advantages and disadvantages.As three common AGV path planning algorithms,the basic principles of A star algorithm,ant colony algorithm and reinforcement learning algorithm are discussed.2.For the global path planning of AGV,an A star algorithm based on adjusting the cost function and finding the optimal path in two directions is proposed.The traditional A star algorithm is improved by adjusting the cost function and exploring the optimal path in both directions.The simulation results show that under the same map environment,the improved A star algorithm has smaller map search range,shorter path length and faster running speed than the traditional one.3.For AGV local path planning,a hybrid enhanced ant colony algorithm based on reinforcement learning is proposed.In view of the low efficiency of traditional ant colony algorithm,a route search method is proposed,which introduces the update strategy of UCB value in the upper confidence bound(UCB)algorithm and Q value into ant colony algorithm,improves the heuristic function,improves the penalty principle of obstacle nodes,and integrates the local optimal path.Compared with the traditional ant colony algorithm,the results show that the hybrid enhanced ant colony algorithm based on reinforcement learning can plan the shortest route in local path planning,and the efficiency of path planning is improved.4.For path planning in unknown environment,a convolutional neural network structure is built,and a deep Q network(DQN)algorithm based on convolutional network is proposed.By designing a simulation experiment comparing the traditional Q-learning algorithm with it,it is verified that the DQN algorithm based on the convolution network is effective and the convergence speed of the algorithm is faster.