Date of Award
Doctor of Philosophy in Mechanical Engineering
Mechanical, Industrial and Systems Engineering
A cooperative adaptive learning-based control (CALC) method and a corresponding experience based controller for a group of identical unicycle-type ground vehicles are proposed in this research, through both state feedback and output feedback. Specifically, consider the generalized dynamic model of the unicycle-type vehicle with unstructured system uncertainties, the proposed CALC method is able to drive all vehicle agents in the multi-agent system (MAS) to their respective desired reference trajectories and accurately approximate the unmodeled vehicle dynamics with radial basis function (RBF) neural network (NN) at the same time. Furthermore, it is shown that the approximation of the unknown dynamics, presented by the NN weights, will reach consensus by converging to the optimal value of approximation along the union of reference trajectories, for all vehicle agents in the MAS.
In addition, a high-gain observer is also developed to estimate the generalized velocities of the vehicles, in case that only the vehicle's generalized coordinates are measured and accessible for the proposed control methods. It is shown that the proposed state feedback controllers can be modified using output feedback with the estimated velocities, and the objectives of trajectory tracking and accurate learning can still be achieved.
An important novelty of the proposed adaptive learning algorithm is that it grants every vehicle in the group the ability of locally accurately identifying the vehicle dynamics not only along the trajectory experienced by itself, but also along the union trajectories experienced by all other vehicles as well, for both state feedback and output feedback controllers. Another novelty of this research is that the control methods proposed in this research can be applied to a more generalized vehicle model with fewer constraints and assumptions, compared to other research results for controlling the unicycle-type vehicles using adaptive learning.
Theoretical analysis, as well as simulations, are provided to show the tracking convergence, learning consensus, and accurate approximation of the proposed control methods.
Dong, Xiaonan, "COMPOSITE COOPERATIVE ADAPTIVE LEARNING AND CONTROL FOR MULTI-ROBOT COORDINATION" (2020). Open Access Dissertations. Paper 1182.