On an anytime algorithm for control

We present an algorithm to calculate the control input when the processing resources available are time-varying. The basic idea is to calculate the components of the control input vector sequentially, in order to maximally utilize the available processing resources at every time step. Alternatively, the algorithm can be viewed as using a sequence of increasingly complicated but accurate models of the process to refine the control input. For the LQG case, we provide analytical performance and stability expressions. For more general cases, we present a receding horizon control based implementation and indicate through numerical simulations that the increase in performance due to the proposed algorithm can be significant.