Learning-based Attitude Takeover Control for Noncooperative Space Targets with Unknown Dynamics

In this paper, we address the problem of post-capture combined spacecraft takeover control for a noncooperative space target with partial constraints and target maneuvering. The real-time inertia tensors identification is avoided, while only the I/O data collected during the on-board operation are utilized to learn a Gaussian process regression model to significantly decrease the system uncertainties. In the meantime, the feedback gains are adaptive with the predicted model confidence. Furthermore, detailed controller design procedures and rigorous theoretical proof of all related closed-loop uniform ultimate bounded (UUB) stability guarantees are provided. Numerical simulations are exhibited to validate the effectiveness of the proposed strategy.