Control of sensory perception using stochastic dynamic programming

A new method for controlling sensory perception is presented. The method is based on stochastic dynamic programming and is set in a discrete event control framework. The method consists of two parts: 1) a dynamic programming (DP) algorithm which evaluates all possible orderings of process monitors; and 2) a real-time sensory perception controller (SPC). The results from the dynamic programming are stored in a lookup table which is used by the SPC. We show how different data fusion methods can be incorporated into the DP model and algorithms without major changes. A case study shows how the SPC increases the average recognition rate while keeping the average monitoring cost low.