How to debug a swarm using non-equilibrium statistical physics

Cooperation is vital for the survival of a swarm. No single bird is faster than a jet plane, and no single fish is faster than a speed boat — humans beat individual animals in air, land, and sea. But, when animals cooperate and swarm, they beat us since biblical times. The science of swarm cooperation contains many open questions, awaiting the discovery of new principles in disordered, far-from-equilibrium, multi-agent systems. A powerful approach to studying swarms experimentally is to design them bottom-up. This requires us to manufacture active particles in large numbers, as when it comes to swarms — more is different. Artificial swarms on both the micro-scale and the macro-scale still struggle to reproduce the agility of natural swarms, in particular, their fluidity when crowded. In my talk, I will describe both microscopic and macroscopic examples that bridge this gap, by looking at swarm cooperation through the lens of non-equilibrium statistical physics. By insuring robots remain autonomous and active when either dilute or crowded, I will show how a completely decentralized swarm can perform collective tasks including transport, corralling, phototaxis, and collective unsupervised learning.