MODIFYING ITERATIVE LEARNING CONTROL TO INCREASE TRACKING SPEED BY MARKOV PARAMETER UPDATES

Similar to humans learning through repetitions of a task, iterative learning control (ILC) learns from previous experience executing a desired tracking maneuver. It observes the error in the current run, and adjusts the command to a feedback controller in the next run. Spacecraft applications include repeated maneuvers of scanning sensors. Some human learning tasks try to learn to track a trajectory as fast as possible. This paper aims to expand ILC methods to not only track the desired trajectory, but also learn to execute the trajectory faster. This raises the frequency content of the trajectory and can excite residual modes or parasitic poles, requiring one to create updated models as one learns. Various effective ILC laws use the system Markov parameters as a model. The recommended model update method from data during iterations is to identify the Markov parameters of an observer. This compresses the number of parameters needed, and allows one to construct as many system parameters as needed. The methods developed can produce high tracking accuracy when the trajectory is too fast for feedback control to be effective. It can also be used to keep learning transients small so that constraints are not violated during the learning process.