Multi-Agent Collision Avoidance Based on DRL and ORCA
Xiyan Zhao, Chongchong Wang, Jia Xu, Li Li, Lucian Buşoniu · 2024
This paper proposes a distributed multi-agent collision avoidance model in dynamic and complex environments based on ORCA and DRL. The main work combines data-driven reinforcement learning with model-based knowledge by integrating imitation learning into reinforcement learning and by designing more effective observations and reward functions. Four strategies are compared, and results demonstrate that our method exhibits superior capabilities.