Safe and Efficient Multi-Robot Path Planning for Warehouse Environments

The logistics industry faces a labor shortage due to an aging population, increasing the demand for Autonomous Mobile Robots (AMRs). AMRs offer flexibility and cost-efficiency, adapting to changing warehouse layouts and allowing dynamic scalability based on workload demands. To control multiple autonomous mobile robots, appropriate path planning is required, and this field is widely studied as Multi-Agent Pickup and Delivery (MAPD). While algorithms such as CA* and CBS ensure optimal paths, they struggle with scalability. In contrast, Priority Inheritance with Backtracking (PIBT) offers faster computation with some compromise in optimality, making it suitable for real-world applications. However, practical challenges arise when transitioning from simulations to real robots, including following collisions and the difficulty of achieving accurate localization. To address these issues, we propose two approaches: Elastic-PIBT, which enhances priority inheritance, and Sequential A*, which adds constraints based on intersection time steps with other robots. Experiments using the Ceiling-Vgm localization system with three to four robots demonstrated successful collision avoidance and stable 30-minute continuous operation. The results showed that even with an average following error of 0.107 meters and an average localization error of 0.36 meters, the robots avoided collisions.