Constraint-following control for uncertain mechanical systems: an intelligent multi-agent game-theoretic approach

Practically all performance requirement of systems can be formulated as constraints. In engineering, systems are tasked, via proper control design, to follow the constraints. There are however two major challenges. The first is that systems possess uncertainty, which may be (possibly fast) time-varying. The second is that systems are mostly nonlinear, due to effects such as nonlinear elasticity, etc. To resolve the challenges, a novel performance measure <^> ss is introduced as a dynamic depiction of the constraint-following error. A new control design is proposed based on this measure. The control design procedure includes two phases. The first phase is to design a control scheme, based on a feasible parameter (meaning the parameter only needs to be selected within a certain range), which renders guaranteed performance regardless of the uncertainty. After that, the second phase is to seek an optimal design for the control parameters. This task is resolved by an intelligent multi-agent game-theoretic approach. We consider there are multi-agents in the design process, each addresses a design consideration. However these considerations are opposite: with one's decreases (or increases) will increase (or decrease) the other. Agents are to cooperate with each other intelligently to reach the optimal choice of the design parameters. The design is applied to an active suspension system, and its performance is verified by comparing with linear-quadratic regulator control and sliding mode control methods.