Research on logistics AGV vehicle path planning based on improved Q-learning algorithm

Abstract

Logistics AGV car as an important part in the intelligent manufacturing, the path planning problem by the attention of many scholars. at present, the path planning algorithm based on reinforcement learning exists the problems of slow convergence and the result is not stable, logistics AGV car in order to get a better return function, you need to perform different actions to gain more experience and information. In order to balance exploration and utilization problems, the traditional Q-learning algorithm introduces the probability value of an exploration factor into the action selection strategy of AGV, selects the state-action pair of the largest Q-value function every time, which leads to the system is easy to fall into the local optimal solution, which also slows down the convergence rate of the whole process. And the final action selection results will also have fluctuations. In order to solve this problem, this paper proposes an improved dynamic adjustment of exploration factor ε strategy, that is to choose different exploration factor ε values in different stages of reinforcement learning, which can better solve the contradiction between exploration and utilization. Through simulation and real experiments, it is proved that the convergence speed of the improved reinforcement learning algorithm is faster and the stability of the convergence result is improved.