Task
Baseline (every team): in
quadsim, take off → fly a waypoint/trajectory
mission → land, closing the loop on the estimated state, under a wind disturbance —
your case-study autopilot is the natural starting point.
Intelligent upgrade (choose one): MPC (Weeks 11–12) or RL (Weeks 13–14).
Show, on a metric you declare in advance, that the upgrade beats your classical baseline —
tracking error, disturbance rejection, or constraint satisfaction.
Mission options (pick one, or propose your own by Week 10)
- 1 · Disturbance-rejection hold — hover / waypoint-keeping under gusty wind; upgrade vs PID on position error.
- 2 · Aggressive trajectory tracking — fast figure-eight or racing line; MPC or RL vs LQR on tracking RMSE.
- 3 · Payload delivery — fly to a point, “drop” mass mid-flight (step change), stay stable; robust tuning or DR-trained RL.
- 4 · Constrained corridor — track a path inside tilt/thrust/position limits; MPC on constraint satisfaction.
The course's built-in LinearMPC and CEM-RL need no extra installs;
do-mpc/CasADi and gym-pybullet-drones/SB3 are welcome for teams that
want the industrial toolchain (see INSTALL §4).
Milestones
- Week 10 — one-page proposal inside the case-study report: mission choice, declared metric, planned upgrade
- Week 15 session — 12-minute presentation + live demo (the mission runs in front of the class, from your code, on the spot) + Q&A
- Code + report — submitted before the Week 15 session starts
Assessment criteria
| Criterion | Weight | What earns full marks |
|---|---|---|
| Working controller & mission | 25% | The baseline mission completes from estimated state under wind, reproduced live and by re-running the code. |
| Improvement over baseline | 25% | The declared metric is measured fairly on both systems; the upgrade wins, or a rigorous negative result is honestly analysed (a well-analysed “it didn't help, and here is exactly why” can score highly). |
| Technical correctness & depth | 20% | The method is implemented and tuned with understanding — formulation stated, design choices justified, failure modes explored. |
| Report (8–12 pages) | 15% | Model → design → experiments → results, reproducible numbers, labelled figures, honest limitations section. |
| Presentation & live demo (Wk 15) | 10% | Clear 12-minute story; the demo runs; questions answered with understanding; every member speaks. |
| Code quality & reproducibility | 5% | One-command run, README, clean structure; no ground-truth leakage. |
Rules
- Same teams and collaboration/AI policy as the case study; the contribution statement is required again and individual adjustment applies.
- The live demo runs on the presenting team's machine from the submitted code version — practise it.
- Late code/report: −20% per day; nothing accepted after the Week 15 session (the demo cannot be made up).
📤 Submit via the MUST LMS
What
one zip per team (code + README + report + slides)
Filename
MLTE03_Final_Team<K>.zipWhere
MUST LMS → MLTE03 → Final Project dropbox LMS link — TO FILL
Deadline
date / time — TO FILL (before the Week 15 session)