Adaptive Hierarchical Multiple Models to Control Dynamic Systems

Modern control theory offers mathematically rigorous and powerful solutions to many dynamic systems. Its limitations are that the classes of dynamic systems covered by the theory are limited to linear systems or certain well-structured nonlinear systems, and the adaptation of the controllers are often slow if the controller parameters are far away from their desired values. On the other hand, the ability of humans to learn to control complex dynamic systems while adapting to changing environments and objectives arises from many mental strategies we employ. Among them are dividing a complex task into smaller manageable subtasks, making use of ``experience with analogous tasks'', and adapting and combining solutions from similar tasks instead of starting from scratch. Motivated by the above observations, we propose a hierarchical multiple models approach to bridge the gap between humans' superior ability to conduct high-level planning and quick adaptation based on prior experience, and the advances of modern control theory in performing low-level control actions for well-structured systems very effectively. Our team consists of Scientific Systems Company, Inc. (SSCI) as the prime contractor, and Professor Robert Jacobs of University of Rochester as a subcontractor for this project.