基于改进Q学习算法的“货到人”系统AGV路径规划

Research on AGV Path Planning of “Goods-to-Person” System Based on Q-learning

随着智慧物流的发展,AGV在现代化智能仓库中的作用愈发重要,并衍生出了“货到人”这一拣选模式。以kiva仓库为研究背景,针对单AGV的局部路径规划进行研究,对传统Q学习“探索-利用”这一困境进行改进,通过引入反正弦函数动态调整贪婪因子ε,满足AGV在前期探索时的随机性,避免陷入局部最优解;后期减少ε保证AGV寻路的目的性,通过栅格方法建立仿真模型。仿真结果表明,改进后的Q学习能够在仓库中找到最优路径,并且运行时间相较于传统Q学习减少了约28%,有效提升了算法的效率。

With the development of intelligent logistics, AGV plays an increasingly important role in modern intelligent warehouse and gives birth to the picking mode of“goods-to-person”. With kiva warehouse as the research background,this paper studies the local path planning of single AGV and improves the dilemma of traditional Q-learning“exploration and utilization”. By introducing the arcsine function to dynamically adjust the greedy factor epsilon,the randomness of AGV in the early exploration can be avoided. The later reduction ensures the purpose of AGV path finding. The simulation model is established by using the grid method. The simulation results show that the improved Q-learning can find the optimal path in the warehouse, and the running time is reduced by about 28% compared with the traditional Q-learning, which effectively improves the efficiency of the algorithm.