DRL-RVO: Learning to Avoid Collision in Crowd Using Reciprocal Velocity Obstacles

This paper proposes DRL-RVO, a Deep Reinforcement Learning (DRL) framework incorporating the traditional Reciprocal Velocity Obstacles (RVO) for safe navigation. In this framework, the conventional RVO algorithm, which is widely used for dynamic obstacle avoidance in autonomous systems, is enhanced by integrating a DRL model to dynamically adapt to the behaviors of other human participants in the shared environment. Specifically, the weight coefficient of the RVO algorithm is updated using DRL in real time, improving navigation performance in dynamic environments shared with human participants. To evaluate the effectiveness of the proposed DRL-RVO algorithm, a comparative analysis is performed against state-of-the-art algorithms. The experimental results demonstrate the superior performance of the DRL-RVO algorithm in terms of navigation success rate, real-time adaptability, and efficiency, highlighting the potential benefits of integrating DRL into traditional RVO-based methods for autonomous navigation in dynamic environments.