In order to solve the AGV scheduling problem of material transportation in intelligent workshop,taking the shortest total walking distance of AGV replenishment task as the goal,a double-layer coding mode was put forward combining the double standards of path selection and task sequencing.At the same time,in order to avoid the clustering of genes on chromosomes in a small neighborhood,an improved genetic algorithm was proposed,which added a variety of mutation processes.Compared with the traditional genetic algorithm,it enlarged the understanding space and prevented the generation of local optimal solutions.Finally,the environment was modeled and simulated by MATLAB,and compared with the basic genetic algorithm.The experimental results show that the improved algorithm can be used to efficiently and reliably solve the path planning problem of AGV under multi-task target.