Tracking under Uncertainty with Cooperating Objects using MOMDPs

Abstract. This paper presents how Partially Observable Markov Deci-sion Processes (POMDPs) can be used for Cooperating Objects control under uncertainty. POMDPs provide a sound mathematical framework to deal with planning actions when tasks outcomes and perception are uncertain, although their computational complexity have precluded their use for complex problems. However, considering mixed observability can lead to simpler representations of the problem. The basic idea is to as-sume that some of the components in the state are fully observable, which is reasonable in many applications. In this paper, target track-ing by means of a team of mobile Cooperating Objects (for instance, robots) is addressed. Instead of solving an usual POMDP, the problem is solved under the assumption of mixed observability, which will reduce the complexity of the original model. Moreover, a decentralized solution in which the COs have no knowledge about others ’ actions is proposed. In this sense, a implicit coordination will be derived from the sharing information among the COs. Finally, some simulations are presented to show the advantages of the methods proposed and how the mentioned coordination arises by using heterogeneous COs with different behaviors.