Robust learning control method in saturating actuators

Abstract

Anti-reset windup is an important tool in the practical implementation of integral controllers. Integral control can have poor performance due to windup of the integral during transient periods when the actuator is operating at its saturation limit. Anti-reset windup speeds up the departure from the saturation limit by stopping the build-up of the integral during saturation. The most basic form of learning and repetitive control makes use of integral control concepts applied in the repetition domain. Therefore, this paper studies the use of anti-reset windup concepts in learning control. The integral is operating in repetitions, but the system dynamics are in time, making the application nonstandard. Various forms of anti-reset windup are developed for use in learning and repetitive control, and shown to improve performance of the learning process, especially when one does not know enough about the system to obtain well-behaved learning transients. Anti-reset windup is also shown to be helpful in situations where the desired trajectory is not feasible, and in situations where the initial conditions are systematically in error, such as in a robot subject to gravity disturbance.