Multi-robot target pursuit: towards an opportunistic control architecture

Abstract

This paper proposes an opportunistic control architecture to action selection in multi-robot, target pursuit problem. Proposed architecture views the problem of action selection in multi-robot environment as an instance of distributed probabilistic inference over the set of robotic agents' available actions, there by constructing a joint probability distribution from local evidence (e.g. robotic agents' respective views of the problem)and the higher level system task perspective. In present work, no explicit inter-robot communication is necessary, instead, robots attain necessary information such as other group members' as well as target's relative positioning information via communicating with a third party agent, the mediation unit. A novel rating system embedded within the control mechanism enables the system to not only determine robots' own action ranking (e.g. default rating) but also incorporates a tactical rating (e.g. opportunisitc rating) at group level. Performance of the opportunistic controller is evaluated in a multi-robot pursuit scenario so as to determine the utility of the proposed sub-ratings system. The analysis of the system performance has been carried out in two parts viz. presence or absence of mediator and absence of opportunistic sub-rating.