Potential Fields Guided Deep Reinforcement Learning for Optimal Path Planning in a Warehouse

Abstract

Using mobile robots for transportation in a warehouse is becoming more and more common. Compared with human staff, these robots can handle the goods more accurately and more efficiently. Using robots can greatly reduce the operation cost of a warehouse. Optimal path planning for these robots can reduce the transportation time, guarantee the safety of the people in the warehouse, and reduce the goods delivery time and increase daily output. In this paper, we propose an optimal path planning algorithm for the mobile robot using deep reinforcement learning (DRL). Potential fields are employed to guide to collect better quality training data to improve data efficiency. The simulation results have shown that DRL can successfully reach the goal position and avoid collision with the obstacles using the potential fields guided trail-and-error method.