What Goes Bump in the Night: Learning Tactile Control for Vision-Occluded Crowd Navigation

Abstract

Expanding the deployment of robots in social environments necessitates safe navigation in contact-prone settings. While collision-free navigation is well-studied, incorporating safe contacts remains underexplored. Traditional approaches mandate robots to freeze upon detecting imminent collisions, risking harm and impeding movement in dense crowds. To address this, we propose a learning-based motion planner and control scheme for omnidirectional mobile robots to navigate ultra-dense social environments using safe contacts. Our approach considers minimizing discomfort to humans, formulated as a multi-task reinforcement learning problem. Evaluated over 160 simulations with varying crowd densities, our navigation scheme achieves a 100% safety factor in crowds up to 1.0 people per square meter and 90% in denser crowds, surpassing previous reported capabilities.

Kyle Morgenstein, Junfeng Jiao, Luis Sentis