Emergence of Adaptive Collective Evasion Through Density-Based Interactions in Swarm Robotics

Abstract

Collective evasion is critical in swarm robotics, enabling groups of simple robots to effectively elude external threats. Previous work on anti-predator behaviors relied on rigid evasion patterns coupled with velocity alignment, which lack adaptability when confronted with sustained and varied predatory attacks and further degrade anti-predation performance. Achieving adaptive collective evasion without predefined evasion rules and explicit velocity alignment is underexplored in swarm robotics. Here, we propose a density-based swarm model for adaptive collective evasion driven by field contextual sensing derived from positional information. By leveraging instantaneous density gradients and temporal fluctuations in density field, our approach enables the emergence of the fountain maneuver, wherein the group exhibits rapid spreading under predation followed by regrouping behind the threat. Compared to coordinated turning and rigid fountain maneuvers, our approach demonstrates better anti-predation performance through more dynamic and longer sustained fountain maneuvers under continuous attacks in both simulations and Human-In-The-Loop predation experiments with real robots, highlighting the practical advantages of adaptive collective evasion in swarm robotics.