| This dissertation extends the fields of learning and repetitive control by developing algorithms appropriate for use in decentralized control systems, such as robot trajectory control. The new field of learning and repetitive control develops controllers that learn to improve their performance at executing a given task, based on experience performing this specific task. Here, algorithms are developed for: (i) P-Integrator based decentralized learning control, (ii) Decentralized indirect learning control based on indirect adaptive control concepts, (iii) Decentralized indirect repetitive control, (iv) Direct model reference learning control and repetitive control. It is envisioned that the system under consideration has a feedback controller in operation executing the desired command, and it is the job of the learning or repetitive controller to eliminate any errors that remain. The development starts with decentralized discrete time systems, and progresses to decentralized continuous time systems allowing time varying coefficients and repetitive disturbances. The algorithms developed guarantee convergence to zero tracking error as the repetitions of the command progress in the case of the discrete time systems, and guarantee the same if the sample time is sufficiently small in the case of the continuous time systems. The original motivation of the learning and repetitive control field was learning in robots doing repetitive tasks such as on an assembly line. The last model applies to the robot problem when the learning process starts with a trajectory in a linear range about the desired one due to the feedback control, and stays within a linear range about the desired trajectory during the learning process. Since the standard industrial feedback controllers are decentralized, it is important to develop learning and repetitive controllers that can be applied in a decentralized manner, as is done in this dissertation. |
