Navigating crowded environments while intentionally interacting with humans (“contact-based” social navigation) remains largely unexplored, contrasting with extensively studied collision-free social navigation. Traditional approaches demand robots to halt abruptly upon detecting imminent collisions, risking harm or impeding movement in dense crowds. To facilitate meaningful robot integration in bustling social spaces, we propose a learning-based motion planner and control scheme enabling safe contact navigation for omnidirectional mobile robots. Evaluating over 360 simulations with varying crowd densities, our approach demonstrates the ability to navigate safely in denser crowds than previously reported.
