Task inference and distributed task management in the Centibots robotic system

We describe the Centibots system, a very large scale distributed robotic system, consisting of more than 100 robots, that has been successfully deployed in large, unknown indoor environments, over extended periods of time (i.e., durations corresponding to several power cycles). Unlike most multiagent systems, the set of tasks about which teams must collaborate is not given a priori. We first describe a task inference algorithm that identifies potential team commitments that collectively balance constraints such as reachability, sensor coverage, and communication access. We then describe a dispatch algorithm for task distribution and management that assigns resources depending on either task density or replacement requirements stemming from failures or power shortages. The targeted deployment environments are expected to lack a supporting communication infrastructure; robots manage their own network and reason about the concomitant localization constraints necessary to maintain team communication. Finally, we present quantitative results in terms of a "search and rescue problem" and discuss the team-oriented aspects of the system in the context of prevailing theories of multiagent collaboration.