Design of a Guidance, Navigation and Control System for a Smart On-Orbit Servicer Mission (SOOS-1)

This thesis presents the preliminary design, implementation, and simulation of a monocular visionbased autonomous Guidance, Navigation, and Control system for Smart On-Orbit Servicing missions targeting defunct satellites in geostationary and graveyard orbits. First, a YOLOv8 object detection model was trained to identify satellite components and tested on a synthetic video frames simulated to model video captured by a monocular camera. These detections were then fed into a depth estimation process using the MiDaS neural network, which generates per-pixel depth maps. Depth readings are extracted from bounding box centers and, together with image coordinates, are used to estimate the relative 3D pose of the target satellite.

To address noise and fluctuations from individual frame detections, a Savitzky-Golay filter is applied to smooth the estimated trajectory over time. Using Two-Line Element orbital data, the simulation initializes five defunct satellite targets and a chaser spacecraft 2 kilometers below each target’s orbit. The Hill-Clohessy-Wiltshire equations are employed for modeling the relative dynamics between the chaser and target satellites. A proportional-derivative controller is implemented in the Hill frame to autonomously steer the chaser into a rendezvous with the target, achieving relative positions within 5 meters and final velocity errors below 0.05 m/s.

Across all five targets, rendezvous was successfully achieved, with maneuver durations ranging from 48,500 to 49,100 seconds. The object detection model reached a mean average precision (mAP) of 97%, while Z-depth outputs ranged from 17 to 21 meters in the test video, validating the feasibility of monocular-based pose estimation. Although the system lacks ground-truth validation for absolute depth, results show promising accuracy in visual tracking and dynamic control. This work contributes a software-level feasibility for SOOS missions, with potential for additional hardware testing and conducting ground truth testing.