Relaxing the Limitations of the Optimal Reciprocal Collision Avoidance Algorithm for Mobile Robots in Crowds

作者：Zhihao Liu, Wenjie Na, Chenpeng Yao, Chengju Liu, Qijun Chen · 发表于：IEEE Robotics and Automation Letters · 年份：2024 · DOI：10.1109/lra.2024.3390561 · 被引用次数：15 · 研究领域：Autonomous Vehicle Technology and Safety、Robotic Path Planning Algorithms、Reinforcement Learning in Robotics

The Optimal Reciprocal Collision Avoidance (ORCA) algorithm is widely used for modeling agents in collision avoidance scenarios. However, suffering from limitations such as the improper reciprocal assumption that each agent is supposed to take half the responsibility for collision avoidance, the performance of ORCA-based mobile robots in crowds is not ideal. In this paper, to relax these limitations, we firstly simplify the planning process of ORCA from the principle horizon to solve ORCA being unsolvable in some cases. Then the escape velocity and collision avoidance responsibility are explored simultaneously based on deep reinforcement learning (DRL) to solve the limitation of local optimum caused by only exploring the responsibility in other works. We compare our method with baselines in environments with different numbers of pedestrians and test in different real-world scenarios. The results show that our method is beneficial in reducing the collision probability and the average number of ORCA no solutions for the robot in crowds.