Probabilistic Projection and Belief Update in the pGOLOG Framework

High-level controllers that operate robots in dynamic, uncertain domains are concerned with two reasoning tasks dealing with the effects of noisy sensors and effectors. They must be able to a) project the outcome of a candidate plan and b) update their belief during execution. In this paper, we show how both tasks can be achieved within the pGOLOG framework [9]. Our approach relies on the idea to characterize the robot’s sensors and effectors as programs written in the probabilistic action language pGOLOG. We are then able to reason about the interaction of the high-level controller and the sensors and effectors through simulation of the concurrent execution of the high-level plan and the pGOLOG model of the sensors and effectors.