| In the face of large-scale modeling errors and uncertainties, the single-model based classical control approach requires high observer-controller gains in order to meet desired tracking objectives. In practice, the sensitivity of the control system to noise, disturbance and uncertainty increases with high gains, causing high-frequency chattering and fast switching, which might make classical design not implementable since available control efforts are usually limited. In addition, high gains could amplify the input and output disturbance in an unusual way, which may excite hidden unmodeled dynamics, resulting in poor tracking performance.;In order to reduce the control gains/efforts as well as to improve the robustness of classical control systems with respect to these non-ideal operating conditions, this thesis introduces a new hybrid adaptive control strategy based on multiple-parameter models/controls. The hybrid strategy reduces the observer-controller gains by reducing modeling errors and uncertainties via identifying an appropriate controller from a finite set of candidates that closely estimate the plant at each instant of time. The proposed scheme works as follows: first, the compact set of parameters is split into smaller subsets, and a candidate controller for each of these smaller subsets is designed; then, the stability-guaranteed Lyapunov-based switching mechanism is employed to select a candidate controller that best approximates the plant at any instant of time among the finite set of candidates. Simulation and experimental studies of the proposed methods are conducted on real robot systems to demonstrate the theoretical development for real-life applications. |
