Unmanned underground parking lots uses AGV systems to move vehicles. The problems of AGV systems include dispatching and routing. In particular, for unmanned underground parking lots, the tasks that AGV needs to perform mainly include car-parking and car pick-up. At the same time, each route must be collision avoidant. This paper studies the dispatching and routing problem(DRP) for AGV systems in unmanned underground parking lots and proposes an AGV dispatching method based on hybrid genetic algorithm and a path optimization method based on time-space graph search algorithm. Results of experiments prove that the proposed methods are feasible and efficient.