AGV Path Planning Model based on Reinforcement Learning

Abstract

With the rapid growth of logistics transportation, automated guided vehicle (AGV) technology has developed speedily. Path planning is one of the key research topics of AGV. It is difficult to plan an optimal path from starting position to target position for AGV in the complex environment. In this paper, reinforcement learning technology is introduced to solve the problem that it is difficult to model AGV path planning due to complex and unknown environment. The Sarsa algorithm based on simulated annealing strategy can effectively guide AGV to plan the optimal path in the grid graph, and improve the success rate. Aiming at the problem that the traditional reinforcement learning algorithm processes data insufficiently in case of large-scale state space, the potential field method combined with deep q-network algorithm is proposed for AGV path planning. The algorithm can effectively guide AGV to carry out optimal path planning, and solve the problem that the traditional reinforcement learning algorithm can not deal with complex space. Finally, these algorithms are applied to the AGV path planning system to simulate the motion state of a single AGV from the loading point to the unloading point. It verifies that our algorithms can effectively implement the AGV intelligent path planning and improve the efficiency of warehousing logistics.