Application of generalized predictive control with learning-based disturbance compensator in repetitive operations

This paper presents the implementation of an enhanced generalized predictive control (GPC) scheme on a two-link planar robotic manipulator performing some repetitive tracking task. The proposed GPC-Gain incorporates a disturbance compensation scheme that combines iterative learning control (ILC) and real-time feedback (RFC) controls. A least mean square error (LMSE) estimator is used to estimate output error caused by repeating disturbances. Through repetitive learning from this filtered output error, ILC predicts the pattern of repeating disturbance. On the other hand, RFC deduces the effect of non-repeating disturbance based on the error feedback information during the ongoing cycle. The learning activity by ILC is regulated using a gain adaptation method. The effect of these estimated disturbances is then compensated in advance in the constrained GPC optimization procedure. Over ten disturbance scenarios, the proposed GPC-Gain scheme reduces the trajectory tracking errors significantly where the average mean squared error (MSE) is merely 49.53% of that of the benchmark. Most importantly, the proposed controller provides a smooth and bounded solution.