Reinforcement Learning-Based Multi-Robot Path Planning and Congestion Management in Warehouse Order Picking

This paper addresses the multi-robot path planning problem in a warehouse environment using reinforcement learning. The warehouse layout comprises of a grid map with multiple robots for retrieval and delivery of orders, inventory pods for storage, and pick stations for receiving outbound orders. The robots are required to pick and deliver orders from target shelves to their corresponding pick stations by navigating in a complex network of aisles. Q-learning algorithm computes optimal paths for the robots, while avoiding congestion in the aisles. Simulation results demonstrate the efficacy of the proposed method in optimizing both travel time and travel distance, thus enhancing the overall operational efficiency of the warehouse.