MS Thesis Proposal
Benjamin Jung
(Faculty advisor: Professor Brian Gunter)
"Data Driven Model Predictive Control for Characterization of a Cold-gas Lunar Drone"
Friday, August 7
12:30 p.m.
Online
Abstract:
This thesis develops and proposes a data-informed model predictive control (MPC) framework
for use in characterizing and controlling the cold-gas propulsion system of a built lunar survey
drone project. The unique advantage MPC offers is not simply just constraint handling or
trajectory optimization; it is it’s ability to run the controller on a detailed, system-specific
prediction model derived with prior simulation and experimental data. Unlike proportional–
integral–derivative (PID) control, which produces a control effort without the need to explicitly
represent the actuator physics, MPC evaluates candidate control sequences through a model
containing the system variables of command history, pressure, temperature, valve and plumbing
transients, thrust, force–torque production, and vehicle motion. Of course, this leads to increased
modeling and computational effort, but these consequences can be accepted in exchange for a
more accurate and control-relevant characterization of the propulsion system.
The research gap here lies at the intersection of cold-gas propulsion characterization and
data-informed predictive control. Current Cold-gas propulsion literature is focused on thruster
design, static or pulsed performance, reaction-control applications, and component-level fluid
modeling. Aerospace MPC literature also operates off of known or separately identified plant
models and applying MPC to guidance, allocation, or constraint enforcement. Although MPC
has been demonstrated with cold-gas spacecraft systems, those systems employ rigid-body and
actuator models designed for planar, orbital, or impulsive control. The lunar drone system
evaluted for this thesis instead uses cold-gas thrusters as primary actuators for sustained lift and
coupled six-degree-of-freedom flight. For this regime, actuator-model error directly affects hover,
attitude, feasibility, and propellant consumption.
The controller will develop from data that already exist for the built drone, including hysteresis
measurements and flight data collected under a PID controller. These records are valuable
because they contain information about the actual actuator nonlinearities and the closed-loop
command-to-motion behavior already observed on the hardware. With this data, information
can be obtained about the initial data-driven model structure, hysteresis and command-history
features, and determination of the operating envelope that the MPC model must represent.
During June, July, and August of 2026, an existing six-degree-of-freedom Robot Operating
System (ROS) simulation will be used to extend this process through controlled excitation, model
comparison, MPC integration, and baseline testing. During Fall 2026, additional hardware tests
will contribute synchronized measurements of command, thrust, pressure, temperature, and
timing to refine the model. The final evaluation will be a validated process that converts existing
PID-flight data, hysteresis data, simulation data, and new hardware data into an hierarchal
MPC prediction model to determine whether that characterization improves factors such as
prediction, hover accuracy, robustness, constraint satisfaction, and propellant efficiency enough
to justify its computational cost.
Committee:
Dr. Brian Gunter (advisor), School of Aerospace Engineering
Dr.Yashwanth Nakka, School of Aerospace Engineering