Robust adaptive visual servoing control for space tether system approach and capture of space debris integrating deep learning perception

In response to the increasingly severe threat of non-cooperative space debris, active on-orbit servicing technology has emerged as a critical measure for ensuring space security. The space tether system, with its advantage of a large capture range and high flexibility, demonstrates significant potential in debris capture missions. However, during the process of approaching and capturing space debris, the uncertainty in target motion, the demands for real-time and precise visual perception, and the field-of-view constraints of the sub-satellite pose substantial challenges to control system design. To address these issues, this paper proposes a visual servoing scheme for space tether system to capture space debris, integrating deep learning-based visual perception with robust adaptive control. The paper first develops a visual perception front-end that combines object detection and keypoint regression models, enhancing its on-orbit robustness through a domain adaptation strategy to provide fast and precise feature point feedback. Subsequently, a visual servoing dynamic model for the space tether system, coupled with camera projection geometry, is established. Targeting this model, a robust adaptive neural network controller is designed based on Lyapunov theory. This controller not only compensates for the flexible vibrational disturbance of the tether online but also incorporates a Barrier Lyapunov Function to explicitly integrate the camera’s field-of-view constraints into the controller design, thereby strictly guaranteeing the stability and reliability of the entire approach and capture mission.