A Bayesian decision theoretic approach for adaptive goal-directed sensing

A general mathematical framework is described for applying Bayesian decision theory to selecting optimal sensing actions for achieving a given sensory goal. The information utilized in the selection process for achieving the goal includes all sensory data acquired prior to the currently considered action. This enables the selection of intelligent sensing strategies to be both adaptive and goal-directed. The authors first show how Bayesian decision theory can facilitate the selection of plans for collecting information relevant to a given task. The approach taken is quite general. It is directly applicable to multi-sensor systems. It could be used in selecting sensory actions to acquire multiple types of information at once. The use of the approach is demonstrated with an example from the domain of robot vision