| Luca Schenato1, Member, IEEE Bruno Sinopoli2, Member, IEEE, Massimo Franceschetti3, Member, IEEE Kameshwar Poolla2, and Shankar Sastry2, Fellow, IEEE
This paper addresses the challenges of control and estimation over unreliable networks, where data transmission and control commands may be lost or delayed. The authors present a generalization of classical control theory to account for the stochastic nature of packet loss and delay. They consider two types of communication protocols: TCP-like (acknowledged packets) and UDP-like (no packet acknowledgment). For TCP-like protocols, the separation principle holds, allowing independent design of the estimator and controller, and the optimal controller is linear in the state. In contrast, for UDP-like protocols, the optimal controller is generally nonlinear. The paper provides conditions for the stability of the closed-loop system and identifies critical probabilities for packet loss that determine the system's stability. It also discusses the design of optimal LQG controllers and the impact of packet loss on their performance. The results are illustrated through examples and theoretical analyses, highlighting the differences between TCP-like and UDP-like protocols in terms of packet loss and control performance.This paper addresses the challenges of control and estimation over unreliable networks, where data transmission and control commands may be lost or delayed. The authors present a generalization of classical control theory to account for the stochastic nature of packet loss and delay. They consider two types of communication protocols: TCP-like (acknowledged packets) and UDP-like (no packet acknowledgment). For TCP-like protocols, the separation principle holds, allowing independent design of the estimator and controller, and the optimal controller is linear in the state. In contrast, for UDP-like protocols, the optimal controller is generally nonlinear. The paper provides conditions for the stability of the closed-loop system and identifies critical probabilities for packet loss that determine the system's stability. It also discusses the design of optimal LQG controllers and the impact of packet loss on their performance. The results are illustrated through examples and theoretical analyses, highlighting the differences between TCP-like and UDP-like protocols in terms of packet loss and control performance.