Trajectory-Preserving Multi-Robot Navigation in Crowded Social Spaces with Online Cooperation Inference

Abstract

We present a decentralized speed regulation framework that augments arbitrary nominal planners with a lightweight cooperation-aware decision layer and integrate it with an existing control algorithm. Each robot runs scalar nonlinear opinion dynamics at a faster time scale than physical motion to infer the cooperativeness of nearby agents from observed speed changes and decide whether to proceed or yield, without solving optimization problems or requiring communication. The resulting opinion state sets a reference speed that can be tracked by any low-level controller. Excitatory-inhibitory interactions encode neighbor cooperation, conflict urgency, and spatial safety. The cooperation identification layer is fully decoupled from the control layer and can integrate independently into other planning stacks. The method is computationally efficient, scalable to large robot populations, and robust in crowded environments with limited onboard sensing. Through extensive simulations and hardware experiments with non-cooperative robots, we demonstrate safe, deadlock-free operation with improved traversal efficiency compared to state-of-the-art multi-robot navigation baselines.