Utilizing Internal State in Multi-Robot Coordination Tasks

The success of a task-achieving multi-robot system (MRS)depends on effective coordination mechanisms to mediatethe robots’ interactions in such a way that a given task isachieved. In the MRS community, many elegant coordina-tionmechanismshavebeenempiricallydemonstrated. How-ever, there is a lack of systematic procedures for the synthe-sis of coordinated MRS. In this paper, we address this issueby presenting a principled framework suitable for describ-ing and reasoning about the intertwined entities involved inany task-achievingMRS Œ the task environment,task deni-tion, andthecapabilitiesof therobotsthemselves. Using thisframework, we present a systematic procedure by which tosynthesize controllers for robots in a MRS such that a givensequential task is correctly executed. The MRS is composedof homogeneousrobots capable of maintaining internal statebut not capable of direct inter-robot communication. Thissystematic approach to the synthesis of coordinated MRSpermits formal identication of the benets and limitationsof MRS composed of robots maintaining internal state andwhen other types of controllers may become necessary, suchas those using communication.The most relevant related work includes work on ndingoptimal policies in partially observable stochastic domains(Cassandra, Kaelbling, & Littman 1994). In MRS, the studyof information invariants (Donald 1995) addresses the prob-lem of determining information requirements for perform-ing robot tasks. The work presented here builds upon ourprevious results on automated synthesis of controllers usinginternalstate (Jones& Matari·c 2003)byaddressingtheissueof uncertainty in sensing and action. Our work presented in(Jones & Mataric· 2004) presents results on the synthesis ofstateless controllers which are communicative but stateless.In the multi-agent systems community, the issues addressedin (Pynadath & Tambe 2002) and (Stone & Veloso 1999)are related to ours in that they are systematically studyingthe role different control characteristics play in coordinatedsystems.