Multi-Robot Learning in an Inherently Cooperative Task

An important need in multi-robot systems is the development of me hanisms that enable robot teams to autonomously generate ooperative behaviors. This paper rst brie y presents the Cooperative Multi-robot Observation of Mul-tiple Moving Targets (CMOMMT) appli ation as a ri h domain for studying the issues of multi-robot learning of new behaviors. We dis uss the results of our hand-generated algorithm for CMOMMT, and then des ribe our re-sear h in generating multi-robot learning te hniques for the CMOMMT appli ation, omparing the results to the hand-generated solutions. Our results show that, while the learning approa h performs better than random, naive approa hes, mu h room still remains to mat h the results obtained from the hand-generated approa h. The ultimate goal of this resear h is to develop te hniques for multi-robot learning and adaptation that will generalize to ooper-ative robot appli ations in many domains, thus fa ilitating the pra ti al use of multi-robot teams in a wide variety of real-world appli ations