Visual Servoing for Final Approach Phase of Spacecraft Proximity Operations with Unknown Targets

A visual navigation and control architecture for the final-approach phase of autonomous spacecraft rendezvous and proximity operations (RPO) with unknown targets is presented. Two spacecraft are considered: a maneuvering chaser, and a possibly tumbling non-cooperative target. The target spacecraft is assumed to be unknown, therefore none of its physical characteristics (geometry, moments of inertia, etc.) are available to the chaser a priori. An image-based visual servo (IBVS) controller is implemented to compute maneuvers for the chaser to achieve a desired goal pose with respect to the target. The controller seeks to minimize pixel location error of matched image feature points on the target. Importantly, because control error is calculated in the image space, a full relative pose estimate between the chaser and target is not required. In order to assess the controller under realistic conditions, the stochastic nature of image feature tracking is modeled via four types of errors: image feature location error, feature nonidentification (occlusion), feature misidentification (outliers), and feature depth estimation error. A bank of Kalman filters are used to track feature points and residual monitoring is used to reject feature measurement outliers. Two novel methods are proposed to generate goal images needed to implement IBVS with unknown targets. The architecture is implemented in a 6DOF simulation using nonlinear relative spacecraft dynamics and camera modeling. Results show that this architecture can robustly control complex RPO maneuvers, to include RPO with tumbling targets.