Model predictive sensor scheduling

The sensor scheduling is to select a sensor (or a group of sensors) from multiple sensors at each time step so as to perform optimally a task based on the sensed data. In this paper, we pose a model predictive type deterministic/stochastic sensor scheduling problem for discrete-time linear Gaussian time-varying systems, and develop an approach to solve these problems based on the dynamic programming recursion. We show first that, in a special case of deterministic scheduling where the Riccati recursion of error covariance satisfies a specific structural condition, the online optimization using the dynamic programming is reduced to a static optimization, so that the model predictive sensor scheduling can be easily implemented online. Next, we discuss the stochastic scheduling problem, and show an alternative condition of optimization reduction, which lead to a stochastic sensor scheduling easily implemented online. Finally, we propose two practical sensor schedulings for deterministic and stochastic case, and discuss an example to illustrate the two sensor schedulings.