Learning-Observer-Based MPC for Autonomous Rendezvous With Noncooperative Targets Under Angle-Only Measurements and Binary Thrust Constraints

Abstract

This article proposes a learning-observer-based model predictive control architecture for the on-orbit cooperative positioning and approaching of noncooperative target. The proposed method specifically addresses the challenges posed by time-varying measurement noise and unknown target maneuvers. In the considered scenario, which relies on angle-only measurements from cooperative positioning satellites and the chaser's single on–off thruster, existing model predictive control methods typically assume full state availability and constant noise statistics for filtering. These assumptions would lead to inaccurate prediction, slow convergence, and steady-state error fluctuations. The proposed framework enables online estimation of the target's relative velocity and unknown acceleration, thereby mitigating system uncertainties and the adverse effects of time-varying noise inherent in angle-only measurements. A multirate sampling structure is designed, where the integration of learning observer estimates improves prediction accuracy and disturbance rejection capability. Engineering constraints, such as the binary nature of the on–off thruster, minimum on/off pulse durations, and thrust vectoring rate limits, are explicitly considered. Simulation results demonstrate that for the approach process from 10 km to 10 m, the proposed method exhibits notable advantages in convergence speed, tracking accuracy, disturbance rejection, and operational constraint compliance compared to conventional approaches.