Résumé : It can be argued that control theory emerged from three engineering objectives: i) stabilize a system, ii) reject unwanted external signals, and iii) minimize the effects of uncertainty. This is the core of the classic servo-regulation problem, where controllers track a reference r while minimizing disturbances d, noise n, and modeling errors.
Over the past two decades, distributed control of networked multi-agent systems, whose central challenge is designing controllers under communication constraints, has received significant attention. At the center of this field is the agreement problem and its celebrated solution, the consensus protocol, which naturally embeds the information exchange graph in the controller structure. This pairs algebraic graph theory with classical control, reducing high-dimensional coordination to what appears like a standard low-order servo problem. Consequently, it has become the de facto standard structure for agreement problems.
In this talk, we argue that there are fundamental differences between this reduction and true servo regulation. We illustrate when and why agreeing systems fail to satisfy the three aforementioned servo objectives. Armed with these insights, we then show that some of these issues can be mitigated by revisiting classical servo techniques. The resulting control structure differs from the consensus protocol but still exploits graph theory to simplify both analysis and design.
