Research on AGV path planning algorithm based on variant beetle antennae search algorithm improved Q-learning

Aiming at the problems of Q-learning (QL) in solving automated guided vehicle(AGV) path planning, such as slow convergence speed in the early stage and easy to fall into local optimum in the late stage, an evolutionary reinforcement learning (BAS-QL) algorithm is proposed to improve QL. There are three main features of BAS-QL. Firstly, a variant of the beetle antennae search algorithm is used to initialize the Q table, which can accelerate the optimization speed in the early stage of AGV path planning. Secondly, use the asymptotic decay Epsilon-Greedy search strategy, using the decay Epsilon value to avoid the algorithm to fall into the local optimum at the later stage to avoid the phenomenon of difficult convergence of the results. Finally, the optimal path of AGV traveling is solved, and the BAS-QL algorithm is verified through experiments. In the n=15 and n=20 raster map comparison experiments, BAS-QL shows the characteristics of short average time consumption, short average distance traveled and low average number of iterations. It shows that the method can also effectively improve the path planning efficiency of AGV while intelligently planning paths.