Optimal supervisory control under partial observation for service robots

Abstract

In this thesis, we deal with optimal supervisory control algorithm for the partial observation system. The objective is minimization of the total cost of the system. The cost is incurred when the supervisor control the event (control cost), or when the event occur(occurrence cost). To attain that objective, we can use the suggested algorithm to design the supervisor for the partial observation case. However, in previous algorithm, the total cost for optimal path could have a overestimation problem, that is, the total cost of the equivalent system is bigger than original system. To solve that overestimation problem, we propose the new optimal supervisory control algorithm and show that the total cost of that proposed algorithm is smaller than original algorithm. With that, by considering the time condition of the system, we propose a new optimal supervisory control algorithm for the system with the time penalty condition. Finally we present the example of the task planning of practical service robot which applying the proposed algorithm.