UQ Space · simulation study · 2026
Airbrake control simulation
200 simulated flights
A Python simulation for airbrakes that trim a rocket's apogee to a target; I wrote it and its controllers for UQ Space. On the 177 of 200 simulated flights of the team's Hera rocket where a 2,900 m target was reachable, three of the four controllers missed it by about 4 m RMS (root mean square) with perfect state knowledge, against 21.4 m for bang-bang. Everything is simulation: no airbrake has been built.
- My role
- Wrote the simulation and the controllers, and reviewed the team's original rocket model
- When
- Oct 2026 – present
- Tools
- Python, RocketPy, OpenRocket, six-degree-of-freedom (6-DOF) Monte Carlo simulation
- Code
- Not published
The full write-up
- The problem
- What I did
- Result
- What went wrong
- Credits
The problem
Airbrakes let a rocket shed speed during its coast so it reaches a target apogee instead of overshooting it. Before building any hardware, the team needed to know which controller to fly and what it would take to fly it well.
What I did
- I fixed the model first: the team's RocketPy model of Hera matched OpenRocket only because several of its errors cancelled. After my fixes I verified it against OpenRocket, and it agrees to within 0.3% on three reference cases. That is model against model, not against a flight.
- I wrote four controllers, target solve, PI (proportional-integral), bang-bang and model predictive control (MPC), and flew each on the same 200 randomised 6-DOF flights at 10 Hz, with actuator lag and the competition's deployment rules.
- 213 test functions in 12 files cover the model and the controllers.
Result
Over the 177 flights where the target was reachable, with the controllers given perfect state knowledge, the RMS miss was 3.7 m for PI, 3.9 m for target solve, 4.0 m for MPC and 21.4 m for bang-bang. The top three are within 0.25 m of each other, so tuning mattered more than the choice of controller.
What went wrong
- No brake data yet: the brake drag is an assumed flat-plate model, because there is no wind-tunnel or computational fluid dynamics (CFD) data, so the brakes' real authority is uncertain.
- The model flies low: real flights of the team's rockets went 2.6 to 5.8% higher than their simulations, so the model needs correcting against flight data before it can predict one.
Credits
Hera's original RocketPy (opens in a new tab) (MIT) model and the team's earlier airbrake hardware are teammates' work. The card shows FHC-1, the airbrake module chosen for the team, which has not been built. The page covers the simulation and controllers I wrote.