Improved Q-Learning Algorithm for AGV Path Optimization

Abstract

With the rapid development of intelligent manufacturing technology, AGV has developed vigorously in the fields of industry, agriculture and scientific research. In recent years, how to plan an optimal path in the navigation system of the automatic guided transport vehicle has become an important topic, which has attracted the attention of scholars. Scholars from different countries have proposed different optimization algorithms for path planning problems. Among them, Q-Learning has made good progress in AGV path planning. Although Q-learning performs well in this aspect, it still has the problem of slow convergence speed and easy to fall into local optimization. To solve the above problems, a Q-learning algorithm based on the beetle antennae search algorithm (BAS-QL) is proposed. In order to improve the convergence speed, the Q table is initialized by using beetle antennae search algorithm. In order to avoid the algorithm falling into local optimum, the attenuated Epsilon value is used. Finally, the optimal path for AGV trolley walking is solved, and the BAS-QL algorithm is verified by experiments. In the n = 8 and n = 10 raster graph experiments, BAS-QL reduces 15.21% and 3.98% in average time, 22.40% and −30.4% in average path length and 77.45% and 43.33% in average iteration times of optimal path compared with Q-Learning algorithm, which shows that this method can effectively improve the efficiency of route planning for AGV.