Research on the Application of Deep Reinforcement Learning in Logistics Scheduling

Abstract

Traditionally, the approaches of logistics scheduling are hard to adapt to the rapid changes of the operational conditions and many of them do not have self-adaptive optimization capability, which will leads to the unsmooths and high operational expenses. In this research, we use Deep Reinforcement Learning (DRL) to improve the logistics scheduling. We take the warehouse order picking as an application example, we build the simulated warehouse environment on a grid system and design the intelligent scheduling based on Proximal Policy Optimization (PPO) algorithm. We compare the performance of it with the usual A* path planning method. The results show that the DRL model built on PPO has better results than the A* algorithm for several important measures, for example, the average steps to finish, the rate of order finish and the total reward accumulation. This investigation proves the possibility of DRL to make flexible and smart logistics scheduling, which is both theoretical reference and .a low-cost simulation tool for the intelligent transformation of logistics enterprises.