Pedestrian-inspired sampling-based multi-robot collision avoidance

Abstract

We present a distributed collision avoidance algorithm for multiple mobile robots that is model-predictive, sampling-based, and intuitive for operation around humans. Unlike purely reactive approaches, the proposed algorithm incorporates arbitrary trajectories as generated by a motion planner running on each navigating robot as well as predicted human trajectories. Our approach, inspired by human navigation in crowded pedestrian environments, draws from the sociology literature on pedestrian interaction. We propose a simple two-phase algorithm in which agents initially cooperate to avoid each other and then initiate civil inattention, thus lessening reactivity and committing to a trajectory. This process entails a pedestrian bargain in which all agents act competently to avoid each other and, once resolution is achieved, to avoid interfering with others' planned trajectories. This approach, being human-inspired, fluidly permits navigational interaction between humans and robots. We report experimental results for the algorithm running on real robots with and without human presence and in simulation.