Application of Anytime Control to an Inherently Unstable Physical System

This paper extends the use of closed-loop anytime control to systems that are inherently unstable in the open-loop. Previous work has shown that anytime control is very effective in compensating for occasional missed deadlines in the computer processor. When misses occur, the control law is truncated or partially executed. However, the previous work assumed that the open-loop system was stable. In this paper, the anytime strategy is applied to an inverted pendulum system. An LQR controller with estimated state feedback is designed and decomposed into two stages. Both stages are implemented most of the time, but in a small percentage of time, only the first stage is applied, with the resulting closed-loop system being unstable for short periods of time. The statistical performance of the closed-loop system is studied using Monte-Carlo simulations. It is seen that, on average, the closed-loop performance is very close to that of the full-order controller as long as the miss rate is relatively small. However, the variance of the response shows much higher dependence on the miss rate, suggesting that the response becomes more unpredictable. At a critical value of miss rate, the closed-loop system is unstable. The critical miss rate found through simulation is seen to correlate well with the results of a deterministic stability analysis. The statistics on the settling time are also studied, and shown to grow longer as the miss rate increases. The transient behavior of the system is studied for a range of initial conditions.