A Multi-Policy Rapidly-Exploring Random Tree Based Mobile Robot Controller for Safe Path Planning in Human-Shared...

Mobile robot path planning in static environments has been extensively studied. However, ensuring a safe path in the presence of stochastically moving humans remains a significant challenge. This work focuses on solving the pathfinding problem of a mobile robot operating in human-shared environments with unknown human motions. To prevent conflicts at the planning level, we propose a multi-policy rapidly exploring random tree (MP-RRT)-based safe pathfinding algorithm. A MP-RRT diverse path generator is developed within this framework to produce multiple diverse candidate paths, which are considered as the initial solution set. Additionally, a dynamic quadrant-based stochastic exploration mechanism is introduced for efficient environment exploration. To obtain an optimally safe path, we design a path optimization mechanism based on stochastic risk evaluation, which explicitly models human motion uncertainties. Finally, an optimal safe path is generated by considering human risks at the planning level to ensure the safety for a robot collaborating with humans. We evaluate the proposed algorithm under different configurations ideal warehouse grid environment from conflict numbers, task success rate, and path reward. The proposed method outperforms A*, MDP, and RRT in terms of conflict number (-70.2%, -72.8%, and -73.8%), task success rate (+66.0%, +95.0%, and +85.7%). Simulation results prove the efficiency of our proposals in safe path planning in human-shared environments.