INTRODUCTION
Self-organizing collective behaviors are found throughout nature, including shoals of fish aggregating to intimidate predators (1), fire ants forming rafts to survive floods (2), and bacteria forming biofilms to share nutrients when they are metabolically stressed (3). Inspired by such systems, researchers in swarm robotics and programmable active matter have used many approaches toward enabling ensembles of simple, independent units to cooperatively accomplish complex tasks (4–6). Both control theoretic and distributed computing approaches have achieved some success, but often rely critically on robots computing and communicating complex state information, requiring relatively sophisticated hardware that can be prohibitive at small scales (7, 8). Alternatively, statistical physics approaches model swarms as systems being driven away from thermal equilibrium by robot…