Reinforcement Learning-Based Formation Control for Networked Fixed-Wing UAVs: Self-Triggered Observer–Feedforward–Feedback Design and Experiment

This article studies the robust optimal formation control problem of networked fixed-wing unmanned aerial vehicles (UAVs) under communication uncertainties and external disturbances. A learning-based observer–feedforward–feedback control framework is constructed. A resilient self-triggered (ST) observer is designed to estimate reference data while enabling intermittent communication under communication uncertainties. By integrating reference estimation with a backstepping technique, the cooperative formation control problem is reformulated as a robust optimal regulation problem. The robust optimal feedforward control law is learned via an off-policy reinforcement learning (RL) algorithm that exploits the collected internal system data and external disturbance inputs. The stability of the constructed closed-loop control system is guaranteed, and Zeno behavior in the ST rule is avoided. The effectiveness of the proposed approach is demonstrated through an experimental study of multiple fixed-wing UAVs.