Controllers as filters: Noise-driven swing-up control based on Maxwell's demon

In this paper we show in simulation that if the controller and the filter are combined into a single computational unit operating like a “Maxwell's demon”, controller-filtered noise can successfully control a system of interest. Using this method, we perform Monte Carlo tests for the swing-up control of the cart pendulum, acrobot and pendubot systems. Results show that filtered noise can indeed act as a swing-up controller, leading these systems to configurations where a locally stabilizing controller can complete the stabilization process. Potential applications of this work include the development of reliable, never-failing human-machine interfaces for rehabilitation, training and skill evaluation.