Design and Analysis of a Estimation and Control Pipeline for a Robot Manipulator in Space

Abstract

Many satellites are rapidly reaching the end of their lifespans, and risk de-orbiting if no action is taken. One common problem satellites face towards the end of their lifespans is that they are running out of fuel, therefore new propulsion units must be delivered in-orbit. Attempting to deliver propulsion units using human astronauts is both dangerous and cost-prohibitive. To meet this challenge, corporate entities such as Northrop Grumman have led efforts to develop cost-effective, robotic in-orbit satellite servicing vehicles capable of delivering life extension payloads to satellites in need of maintenance. The realization of robotic deliveries of life extension payload is the beginning of a larger effort to perform more general on-orbit maintenance tasks using robotic tools.

In this thesis we present an integrated tracking, estimation, and control framework for space robots, along with an environment to simulate in-orbit satellite servicing missions. We show that existing methods for operational space control of floating base manipulators can be extended to partially observable environments by incorporating contact force information into the estimation and control problem.

For our control subsystem, we first evaluate multiple control solutions (adaptive model predictive control, nonlinear model predictive control, operational space control). After evaluating both optimization-based approaches, (model predictive control) and classical approaches (operational space control), we selected an operational space control method to send control commands. We selected this controller because we concluded that optimization-based methods struggle to run in real time for our high degree of freedom system. We present two types of analysis for our controller: feasibility, and stability analysis. We perform multiple types of analysis to demonstrate the efficacy of our controller. We perform feasibility analysis in order to numerically compute the space of initial robot end-effector poses for which the controller can successfully complete a docking mission. We perform stability analysis to guarantee the stability of our controller using analytical methods from non-linear control theory.

We analyze our vision tracking subsystem by conducting a series of experiments to understand the configurations in which our vision subsystem succeeds and fails, and with what degree of confidence it reports measurements of the client satellite. We focus on understanding how the motion of our robot arm and the end-effector payload effects the performance of the tracker through occlusions of the client satellite, and we search for configurations of our robot base and arm which minimize occlusions. This analysis enables us to better position the arm and the MEP during docking operations.