Local Estimation vs Global Information: the Benefits of Slower Timescales.

Self-organized partitioning into subgroups is often a desirable skill of a swarm system, as seen in examples like honeybee swarms dividing into task forces or worker robots spreading across different target regions with various demands. Each agent in the swarm selects a target region based on its perception of the world's current state, i.e. worker count and work demand in each region. It's intriguing to assume that the better agents perceive the world, the more efficiently the swarm performs. In this study, we compare a swarm with flawless perception of the world's state, i.e. access to global information, to one where robots estimate the state through local interactions. Unexpectedly, our results reveal that the swarm relying on local estimates outperforms the one with perfect perception. Further exploration into information access timing suggests that a slower timescale of access might play a key role in the swarm's success.