Model Predictive Controller for 3-dof free floating platform with binary actuated thrusters and reaction wheel

Air-bearing-based platforms are used to simulate microgravity on Earth. Therefore, these platforms can be
used to test for example Active Space Debris Removal (ASDR) maneuvers. European Space Agency’s Orbital
Robotics and GNC Lab (ORGL) features a 5m × 9m flat floor. A 220 kg heavy air-bearing platform combined
with a cold gas propulsion system with eight thrusters and a reaction wheel stacks up to a satellite simulator.
The thrusters can only be on/off-actuated while having time constraints on the opening and closing times of the
valves. The existing control scheme based on an offline trajectory optimization and a TVLQR does not respect
these constraints and therefore shows a not satisfying performance on the real system.
The goal of this thesis is the development of an MPC controller for the platform that directly controls the
thrusters, considering the on/off behavior and the timing constraints to enable (thrust-) optimal control.
The first goal was to find a way to express binary input constraints so that a solver is able to find a solution
in real-time. A Mathematical Program with Complementarity Constraints (MPCC), non-convex Quadratic
Program (QP) and Mixed Integer Program (MILP) was compared on the simulation of a simplified, de-coupled
model of the system. Out of the different formulations, the Mixed Integer Linear Program has turned out to be
the only really fast and working formulation.
Therefore, a linear formulation of the real optimization problem including the thruster timing constraints has
been developed. The system dynamics are formulated as the fully coupled model including all eight thrusters
and the reaction wheel. The latter is modeled with its velocity limits and torque constraints.
The controller has been implemented in C++ and tested on two different simulators, which show that it
is working in real-time. Moreover, the controller has been integrated into the real system, where it has been
shown that the controller works even under the disturbances of the not perfectly even floor. It outperforms the
existing TVLQR controller in accuracy and reliability.