Sim to real · Volume 2
Adaptive Quadrocopter
Build a simulator, establish classical baselines, train an adaptive controller, then test it on a printed quadrotor. Nine modules connect the mathematics to measurable experiments. Every result belongs in the report, including the runs that fail.
Start on your laptop. Build the hardware when you’re ready.
Make the theory fly
The same question runs through every build: what must a network learn to adapt to physics it has never seen?
The answer becomes a simulator, a fair benchmark, a learned controller, and a drone you made yourself.
Write and validate a 6-DoF quadrotor simulator with explicit motor, battery, noise and delay models.
Compare PID, LQR, EKF adaptation and learned adaptation with the same reproducible benchmark.
Design a parametric frame, build a measured physics model, and work toward controlled indoor flight.
Deploy a compact policy through a shared controller interface and publish evidence for all nine gates.
The curriculum
Nine modules. Each one has a lesson, a lab, a deliverable, and a measurable gate. Keep the evidence, including the results that surprise you. On the map, each episode companion sits next to its module.
Arrow keys move between modules and episodes.
Before you begin
Read the lesson, implement the lab, and submit your artifacts to the module’s gate. Record the command, result, commit, configuration, and seed in your progress log.
“Studied” is a reading bookmark. A passed gate is a reproducible result. Hardware gates also need a trainer’s review of your logs and video.
There is no fixed time estimate: progress depends on the builds and debugging. The software part needs no hardware purchase or paid cloud compute.
The physical part
The specification budgets approximately €1,160–€1,200 for parts. Treat this as a planning estimate; verify prices, firmware support, and compatibility before ordering.
| Part | Quantity | Budget | Build note |
|---|---|---|---|
| Crazyflie Bolt 1.1 | 1 | €220 | Early Access hardware: validate firmware, ESC protocol, deck support and custom-controller compatibility before purchasing. |
| Lighthouse positioning deck | 1 | €102 | Confirm deck mounting, firmware support and unobstructed optical coverage on the proposed frame. |
| Lighthouse V2 base stations | 2 | €478 | Bitcraze asks for at least two. Priced at the HTC Vive shop; Amazon.de listed them at about €575 in October 2026, and used ones can cost much less. Plan the tracking volume before purchasing, mount rigidly and verify coverage throughout the intended flight envelope. |
| Crazyradio 2.0 | 1 | €46 | Used for the common off-board controller link and telemetry. |
| 1404 motors, approximately 4500 KV | 4 | €64–88 | Candidate 3S/3-inch combination; verify propeller, voltage, current and thrust data for the exact motor. |
| 20 A class 4-in-1 ESC with Bluejay firmware | 1 | €67 | Bluejay reports each motor’s speed over bidirectional DShot. Once you switch the Bolt to DShot, its firmware enables that telemetry by default and checks every motor’s speed before arming, so a stock BLHeli_S ESC fails the check unless you flash it to Bluejay or turn the telemetry off. A 4-in-1 usually mounts on 20×20 mm holes, the Bolt on 30.5×30.5 mm. Power it straight from the battery and run only the signal wires from the Bolt’s ESC ports; Bolt distribution is limited to 8 A per motor. |
| 3S 450 mAh LiPo with XT30 | 3 | €36–39 | Check size, mass and current requirements; include a battery in every all-up mass calculation. The Bolt’s default battery thresholds are 2S values; set 3S thresholds in Module 7. |
| 3-inch propellers | 2 sets | About €10 | Match motor shafts, rotation directions and guards. Keep spares for damaged props. |
| Suitable LiPo charger and charging protection | 1 | About €60 | Use a charger compatible with cell count and chemistry and follow the battery manufacturer’s charging instructions. Store packs at storage voltage and discharge them before disposal. |
| Load cell and HX711 | 1 | €17 | Measures axial thrust directly, and the torque constant through a lever-arm fixture. The HX711 samples at 10 or 80 Hz, too slowly for motor lag. |
| USB microcontroller board for the HX711 | 1 | About €6 | The HX711 needs a reader. A Raspberry Pi Pico or a similar board streams its readings over USB to the laptop, which logs them beside the Bolt’s motor command and battery voltage. |
| PETG or suitable nylon filament | 1 kg | About €25–40 | Validate the printer, nozzle and material process. Slicer settings and actual printed masses belong in the build record. |
| Vane anemometer | 1 | About €25 | Measures the desk fan’s air speed along the flight path for the Module 9 fan condition. |
| Optical tachometer (optional) | 1 | About €20 | Measures motor speed for kf and motor lag in Module 8. With the Bluejay ESC it cross-checks the DShot telemetry, whose readings depend on the configured pole count. Without it, use bidirectional DShot telemetry or the accelerometer method described there, or declare those values unresolved with an assumed range. |
The design targets 210 g all-up, with every test configuration below 250 g. Flight work uses a contained indoor test area, a kill switch, and the staged checks in modules 07–09. Confirm the obligations and insurance that apply where you fly. Parts list with shop links (CSV) ↓
Follow the experiment
Read the written companion to every planned episode. Recording and customer access are forthcoming; the Free and Customer labels show the intended recording release track.
The course’s opening thought experiment: what should a hidden vector learn about a drone?
Read companionA force balance, a quaternion and three checks make the first build inspectable.
Read companionA frozen benchmark turns model mismatch into a reproducible failure.
Read companionThe EKF estimates an effective mass, with a lesson in what hover cannot reveal.
Read companionDesign the ablation that can confirm or overturn the opening hypothesis.
Read companionCausal timestamps and dataset boundaries matter as much as the network.
Read companionMove from privileged training information to a history-only controller.
Read companionA reproducible protocol for comparing JAX and an optional Isaac Lab run.
Read companionParametric CAD becomes an executable description of geometry and mass.
Read companionCommission the aircraft through inspection, bench checks and stock flight.
Read companionMeasure the gap, including the uncertainty in the measuring tools.
Read companionReplay interfaces, test fallbacks and verify inference before the first learned flight.
Read companionThe final comparison includes the possibility that a classical controller wins.
Read companionChoose your depth
All written lessons, companions, and downloads on this site are accessible now. These tracks describe the planned series and its teaching resources.
Free
Customer
Trainer
Your first build is one click away