Collective Sorting with Multiple Robots

Abstract

Inspired by the behavior of social insects, we tackle the problem of sorting objects with a group of robots under the control of reactive behaviors. Our control algorithm is based on earlier studies of this problem, but depends on more sensing than previous minimalist solutions. With additional sensing information and our simple behavioral rules, we empirically demonstrate that our control algorithm is able to create a complete separation of objects of two different classes. Through simulation, we also show the robust convergence of the sorting process, which previous algorithms could not achieve. This result is independent of the number of robots participating in the task, the initial configuration of the world, and the number of objects to be sorted. We also show indirectly that sorting is not a strictly cooperative task in the sense that even a single robot is capable of performing the task, though at a reduced pace. Finally, we present a model that characterizes the growth of object clusters, which can be used to understand the dynamics of the sorting process