A vision based control system for autonomous rendezvous and capture of a Mars Orbital Sample

Abstract

NASA's Mars Sample Return (MSR) mission involves many challenging operations. The current mission scenario utilizes a small Orbiting Sample (OS) satellite, launched from the surface of Mars, which will rendezvous with an Earth Return Vehicle (ERV) in Martian orbit. One of the highest-risk operations is the guidance of the OS into the capture mechanism on the ERV. Since the OS will most likely be passive (with no attitude or propulsion control), the ERV must determine the OS' location in Martian orbit, and maneuver itself to capture it. The objective of this project is to design and develop a vision-based tracking and capture system using the SPHERES test bed. The proposed Mars Orbital Sample Return (MOSR) system uses a SPHERES satellite to emulate the combined motion of the capture satellite and the OS. The key elements of the system are: (1) a modified SPHERES satellite with a white shell to match the optical properties of the OS; (2) a capture mechanism; and (3) an optical tracking system. The system uses cameras mounted on the capture mechanism to optically track the OS. Software on the capture mechanism computes the likely maneuver commands for a capture satellite, which are then translated into relative motions to be performed by a SPHERES satellite, acting as the OS. The focus of this thesis is on the vision-based algorithms and techniques used to ensure accurate 3-DOF ranging of the OS. The requirements of the OS tracking system are severe and require robust tracking performance in challenging illumination conditions without the use of any fiduciary markers(on the OS) to assist as a point of reference. A brief literature survey of common machine vision techniques for generic target tracking (in aerospace and other fields) is presented. In the proposed OS tracking system, two different methods are used for tracking and ranging of the OS. A Hough Transform algorithm is used to ensure accurate tracking of the OS in the 'near' field within all possible illumination regimes. A Luminosity based tracking algorithm is used to track the OS in the 'far' and 'near' field. Results from testing at MIT's Flat Floor Facility are presented to show the performance of these algorithms in an integrated Kalman Filter. Lastly, a new Model Predictive controller design is proposed for the fuel-optimal capture of the OS. Implementation and testing of the controller in the SPHERES satellite is presented and the comparisons to the SPHERES PD control system are revealed to highlight its strengths.