Robot rostering: Coalition formation for long-term missions with work shifts
In the context of coordination and planning in collaborative multi-robot/agent systems, we consider a general reference problem that includes tasks that are spatially localized and have an associated service time, and accounts for the use of a heterogeneous team, in which different robots may have a different performance on the same task. A mixed integer linear formulation is introduced and used to solve the problem model in a centralized iterative manner: in closedloop, team-level plans are adaptively computed and sent out. Unfortunately, a centralized scheme can suffer from computational and communication shortcomings. Therefore, we introduce a top-down recipe for decentralization, aiming to balance the tradeoff among implementation costs, computational requirements, and quality of coordination. The decentralized architecture depends on various aspects that we study through an empirical sensitivity analysis. Results show that the impact and the relationships among the different aspects are far from being obvious or intuitive. A number of practical lessons are learned, that could apply to other similar problems and/or decentralized architectures derived in the same top-down modality.