Inverse Learning Approach to Reward Design for Deadlock-free Route Planning Problems for Automated Guided Vehicles

Abstract

In recent years, with the development of IoT technologies such as 5G, it is expected to operate large-scale route planning for many AGVs (Automated Guided Vehicles). Deadlock is caused on the routing of AGVs when the route us conducted by conventional reinforcement learning approach. In this study, we propose a route planning method that avoids deadlocks by using inverse reinforcement learning to design a reward that is close to the route planning optimization method. We compare the proposed method with conventional deadlock avoidance methods to verify the usefulness of the proposed method.