Reinforcement Learning‐Based Event‐Triggered Resilient Coordination Control for Spacecraft Formation Flying

ABSTRACT

This paper studies resilient coordination control for a spacecraft formation subject to limited communication resources, admissible communication‐link faults, actuator faults, and uncertain nonlinear dynamics. A distributed data‐driven coordination scheme is developed using a resilient observer, event‐triggered communication mechanisms, and a learning‐based fault‐tolerant controller. The off‐policy reinforcement‐learning procedure obtains a finite‐dimensional neural‐network approximation of the nominal optimal control policy. An adaptive compensator mitigates the bounded actuator‐fault‐induced input uncertainty considered in this paper. A self‐triggered mechanism based on locally stored and received fault‐affected interaction information avoids continuous neighbor monitoring. Asymptotic coordination and Zeno exclusion are established for the ideal closed‐loop system under the exact HJB policy and the finite‐dimensional learned policy is evaluated numerically. A spacecraft formation flying case study reports learning diagnostics, quantitative tracking and communication metrics, component ablations, fault sensitivity tests, and actuator‐level compensation results.