Optimal Design for Assured Performance of Interactive Multibody Systems: Guaranteed Controls for Multi-pursuers, Estimation, Optimal Learning, Scalable Uncertainty Sampling, and Time-critical Communic
Abstract
Performance guarantees for single agent dynamics have been developed since the 1960s and give flight controller designs with assured gain and phase margins, and other robustness properties. However, modern day systems are often composed of multiple interacting agents, or multibody systems (MBS), and a large gap exists between performance guarantees for individual systems andthose for the overall MBS. When multiple dynamical systems interact on communication networks, VFR aircraft in formations, or through other dynamical coupling mechanisms, guaranteed performance of each individual system cannot guarantee performance of the overall MBS, which may even fail the simple test of overall stability. Attempts to guarantee provable performance for uncertain networked MBS often rely on expensive Monte Carlo simulations basedon thousands or millions of tests. This research will develop a mathematical foundation for analysis and design of interactive distributed control systems for networked MBS with guaranteed performance and assurances of robustness and stability. Our approach to MBS distributed controls design with performance guarantees is based on three Main Paradigms. First is optimal localdesign at each agent for computable margins and robustness bounds. It is known that optimal local design for multi-agent systems on graphs guarantees global team synchronization and stability. It is known that a robust control Lyapunov function for nonlinear systems is equivalent to existence of a value function for a suitable optimal control problem. Second is a distributed stimation and LQG/LTR-like approach to address unknowns and a scalable and highly efficient uncertainty evaluation alternative to Monte Carlo simulation for complex intercommunication MBS. Third is a graph-theoretic intelligent communication and data measurement approach to meet time-critical missions. Most existing cooperative control laws for networked multi-agent systems rely on nonoptimalfeedback of locally defined error signals in terms of the states or outputs of an agent and its neighbors. Being non-optimal, performance and robustness guarantees cannot be given. The Technical Approach is structured into three Scientific Objectives. First is guaranteed performance of MBS control that is based on locally optimal design, and guaranteed adaptive methods for learning optimal solutions online in real-time. Second is guaranteed performance control in MBS with uncertainties and unknowns using distributed estimation and non-Monte Carlo sampling techniques. Third is multi-pursuer/evader games as a case study of MBS, andgraph-theoretic intelligent communication and data measurement techniques to further improve the performance to meet time-critical missions. Anticipated Results include:(1) A new theoretic foundation of guaranteed performance control for MBS that is extended from performance guarantees for single agent dynamics, such as linear gain and phase margins, sensitivity, and nonlinear robustness bounds.(2) A guaranteed performance MBS control framework to deal with unknowns and high dimensional uncertainties.(3) Control and decision algorithms for multi-pursuer/evader games as a case study of MBS, that improve the performance of MBS control using graph-theoretic approaches such as intelligent communication and data measurement techniques to meet time-critical missions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 10, 2018
- Source ID
- N000141812221
Entities
People
- Frank Lewis
Organizations
- Office of Naval Research
- United States Navy
- University of Texas at Arlington