Cooperative output regulation, which extends output regulation from single systems to multi-agent systems (MASs), offers a unifying framework that accounts for heterogeneity in agent dynamics, dimensions, and uncertainties, and enables tracking and rejection of a broad class of signals. Typical cooperative control problems, including formation tracking and containment, can be cast as a cooperative output regulation problem (CORP). The goal of the CORP is to design a distributed control law that ensures closed-loop stability and makes the output of every follower asymptotically track (reject) a class of reference (disturbance) signals. This problem has been primarily addressed using two approaches: the distributed observer and the distributed internal model.
Burak organized a workshop on cooperative output regulation with Ahmet Taha Koru (University of Texas at Arlington) and Yamin Yan (Nanyang Technological University) at the American Control Conference, Toronto, ON, 2024. Jie Huang (Chinese University of Hong Kong), a pioneer in the topic, gave the first talk in the workshop.
IJC 21 extends the key solvability result of the conventional internal model approach to heterogeneous MASs both in dynamics and dimensions over general directed graphs. This extension has become key to distributed control laws via the distributed internal model approach, ranging from dynamic state feedback to dynamic output feedback with local measurement and dynamic error feedback. When a central authority is permitted to design such distributed controllers, structured Lyapunov inequalities in IEEE TAC 21 provide convex relaxations to the original nonconvex structured control problem and transient performance guarantees. While this may be suitable for medium-scale, repetitive operations under well-defined conditions, it is impractical for large-scale autonomy in adverse conditions. We, however, derived agent-wise local design methods involving linear matrix inequalities, where each agent solves (1) a local stabilization problem with a disturbance attenuation level determined by a spectral property of the graph (IJC 21); (2) a local robust D-stabilization problem (IEEE TAC 23) for arbitrary switching between graphs or slow switching with an average dwell time constraint (IEEE TAC 24). These methods are scalable, serve as effective preliminaries for real-time onboard distributed control redesign in the case of shifts in mission objectives and operational conditions, and the second set of design methods ensures performance guarantees for the overall MAS (e.g., minimal decay rate and damping ratio) through each agent's local design. The distributed internal model approach is robust (with respect to small agent-level parametric uncertainties) in steady-state tracking response over the distributed observer approach. To address large-scale uncertainties, Book Chapter 24 proposes a distributed dynamic state-feedback control law that consists of a distributed reference model characterizing the desired closed-loop response, a parameter adaptation rule suppressing agent-level linearly parameterized matched nonlinear uncertainties, and a decoupling virtual tracking error. The proposed distributed control law, especially the decoupling virtual tracking error, breaks the original problem into two problems: (1) CORP, where our distributed internal model results are used to construct the distributed reference model; (2) model reference adaptive control for each agent.
Formation Keeping for Cart-Inverted Pendulums and Carts: AIAA SciTech 19
Containment of Multi-Human Multi-Agent Systems Under Time Delays: IEEE TSMCS 24
Output Formation Tracking of Networked Quadrotors and Mobile Robots: AIAA SciTech 26
Multi-Partite Output Regulation with an Application to Line Formation of Networked Mobile Robots: arXiv 25 (conditionally accepted to IEEE TAC)