Jump to content

Robust flight control, for fast air platforms

From Apollyon Wiki
Guidance & control · flight control settings and the simulation behind them · core subsystem

Flight control for fast aircraft that fly hard: high dynamic pressure, sharp high-G turns, and actuators at their limits. It pairs a simulation built from real flight data with dense onboard recording, so the aircraft stays stable where standard linear autopilots lose control.

01 · Physical envelope

Flying where the usual assumptions stop working

Standard autopilot designs work well where flight stays gentle: steady cruise, hover, and shallow turns. Near the edge of what the aircraft can do, that changes. Behaviour turns uncertain and non-linear, small inputs produce large and uneven responses, and much of it is hard to write down in a model beforehand.

Fig. 01steady cruise · near the limits
HOW THE AIRCRAFT BEHAVES AT LOW SPEED AND NEAR ITS LIMITS faster STEADY CRUISE AND HOVER motion is smooth and each axis moves on its own FAST FLIGHT NEAR THE LIMITS motion is coupled and the margins run out expected by a simple model what the flight shows physical limit THRUST thrust rotor speed Thrust rises steadily at first. Near full speed it flattens and leaves the straight line. MOTOR RESPONSE rotor speed time slow to rise, unsteady at the top The motor is slow to answer, then unsteady near full power. AERODYNAMICS drag speed Drag grows with the square of speed — about fifty times from 20 to 140 m/s. ROTATION LOW RATES HIGH RATES others join in R P Y R P Y Turns stay separate at low rates. At high rates each turn drags the other axes with it. SENSING measured pull time ±16 g limit A hard pull plus shaking can push past what the sensor can read. ACTUATION full power cruise near the limit All motors sit near full power, with little room left for a gust. At low speed the motion is smooth and easy to predict. Near the limits it turns coupled and non-linear.
Fig. 01 At low speed the motion is smooth and each axis moves on its own. Near the limits the actuators run out of room and the motion becomes coupled and non-linear.
02 · Limits of linear control

Where the usual controllers run out of room

Most controllers assume there is spare actuator power and that the airflow matches the model. Near the limits, both assumptions fail. The figure below walks through the two standard approaches and the alternative used here: control settings taken from flight tests, with limits that keep demand inside what the aircraft can actually do.

Fig. 02fixed loops · online solver · scheduled law
WHY FIXED CONTROL SETTINGS RUN OUT OF ROOM A · FIXED CASCADE TARGET PATH POSITION ATTITUDE RATE MIXER fills up first MOTORS Control in steps. Each step needs spare power. B · ONLINE OPTIMISATION MODEL + STATE + GOAL equations of the aircraft ONLINE SOLVER answer computed in flight MOTORS Plans ahead, live. Needs exact models. C · SCHEDULED CONTROL WITH LIMITS — USED HERE STATE + AIR DATA + LIMITS what is known, what is allowed SCHEDULED LAW + PROTECTION gains, bounds and distribution FINS + MOTORS Settings from flight tests. Demand stays within limits.
Fig. 02 Fixed cascades need spare power at each step, and live planners need exact models. Neither is guaranteed near the limits, so the settings used here are taken from flight tests and bounded to stay within what the aircraft can do.
03 · Physics backbone

The simulation backbone, built from real flights

Control settings are only as good as the simulation behind them. Standard simulators cover cruise and gentle flight well. Ours is extended toward harder conditions flight by flight, alongside work that stays specific to each airframe.

Each sortie records dense telemetry across the sensor and actuator buses. Afterwards the same inputs are replayed through the model, and where the two disagree the gap is traced to something physical — airframe flex, motor heat, voltage drop — before falling back on generic noise terms.

What the simulation has to get right
AreaWhat is modelledWhat goes wrong without it
AerodynamicsAirflow separating from the surface, shock-driven separation, and uneven downwash between front and rear lifting surfaces.The aircraft can stall without warning in a hard pull-up.
Propulsion & thermalMotor coils heating up, speed controllers cutting back with heat, back-EMF saturation, and battery voltage sagging under peak current.Thrust fades in a long high-G dive and the aircraft loses height.
AeroelasticsFuselage flex, fin flutter, and structural vibration reaching the IMU mount.Control surfaces buzz and servo motors overheat.
Actuation physicsPlay in servo gearboxes, slower response under air load, and deadband around centre.Response lag that can feed oscillation in transonic flight.
A shared baseline, plus work per airframe

Fixes that are not tied to one project are carried over to other projects too. Each airframe still needs its own flight hours and its own settings, but the baseline simulation keeps getting better, making the work on each next airframe less tedious.

Fig. 03train · fly · compare · correct
HOW THE SIMULATOR CATCHES UP, ONE REAL FLIGHT AT A TIME a pilot sets the target · each pass leaves the simulator closer to the aircraft EXPERT FLIGHT a pilot shows what is possible CURRENT SIMULATOR built from earlier flight data TRAINING IN SIMULATION settings tried across many cases CONTROL SETTINGS loaded onto the aircraft FLY IT FOR REAL the aircraft tries the same flight LOGS AGAINST SIMULATION same inputs replayed, gaps noted CORRECT THE SIMULATOR motor curves and airflow fixed repeat one round = train, fly, compare, correct THE LOOP ENDS WHEN THE AIRCRAFT MATCHES THE PILOT AND THE SIMULATOR PREDICTS THE FLIGHT. Each pass leaves the simulator closer to the aircraft, so the next airframe starts from a better base. Tuning sets stay specific to each airframe. What carries over is the method, the simulator core and the test pipeline.
Fig. 03 Each pass trains the settings, flies them, compares the logs and corrects the simulator. The loop ends when the aircraft matches the pilot and the simulator predicts the flight.
04 · Sim-to-real flight loop

What one test flight gives back

Test flights push the airframe to its structural edge, record everything, and the data is worked through within hours. Each campaign comes back with four concrete updates — one each for the simulator, the control settings, the onboard software, and the hardware — and the next airframe starts from there.

What one flight updates
What the flight showedSystem updatedWhat was changed
The actuator moved slower than expected under air loadSimulator physicsThe torque and back-EMF models were corrected to match the recording.
The pitch-up did not rotate the way the model predictedControl configurationGain tables and envelope limits were reset at the newly measured edge.
A timing spike on the sensor busOnboard runtimeKernel scheduling and DMA timing were checked against the recorded traces.
The battery ran hotter than modelled and its resistance rose earlyHardware specificationThe power bus, cell cooling, and wiring were revised.
Fig. 04fly · compare · update · repeat
WHAT ONE TEST FLIGHT GIVES BACK flown near the limits, recorded in full, reviewed within hours FLIGHT NEAR THE LIMITS current, heat, timing and vibration recorded COMPARISON WITH THE SIMULATION flight data laid against the prediction, within hours SIMULATOR air and motor models fixed to match the flight → simulator CONTROL SETTINGS gain tables and limits reset at the new edge → control law ONBOARD SOFTWARE scheduling and timing checked on the recording → onboard runtime HARDWARE power and cooling fixed for the next build → hardware revision THE NEXT AIRFRAME starts from the updated base one round trip takes days WHAT ONE FLIGHT OFTEN SHOWS a motor slower than expected → simulator a turn missing the command → control settings a timing spike on the bus → onboard software a battery hotter than modelled → hardware WHY EACH ROUND GETS CHEAPER The rigs and pipelines are built once. Each later round costs less and returns more. shared base kept by the company flight rounds →
Fig. 04 One flight near the limits updates four places at once: the simulator, the control settings, the onboard software and the hardware. The next airframe starts from that updated base.
05 · Across airframes

One method, many airframes

Each airframe needs its own control settings, worked out in its own flight tests. What carries over is the method: how flights are instrumented, how the simulation is corrected, and how settings are cleared for flight. A new platform goes through the same pipeline instead of starting from nothing.

How each platform uses it
PlatformSpeedHardest control problemController
Ahuti Interceptor400 km/h band
(498 km/h sprint testbed)		Switching from vertical launch to fast forward flight, then steering into the intercept.Own rotorcraft controller · same method
Nightshade ADX-1700 km/hMoving from slow loiter to a fast dash, then holding a steep high-speed dive.Veronte 1x · strike package
HemlockMach 0.7–0.8Low flight over the sea at high subsonic speed, then a steep dive onto the target.Own implementation · same method
Piranha USVUp to 65 km/h (design target)Holding course and trim at high speed in rough seas, across very different low-speed and sprint behaviour.Own marine controller · same method
06 · Used by

Platforms that use this subsystem

Integrated into
Ahuti Interceptor — high-speed interception (main user) Nightshade ADX-1 — terminal dive and loiter stability Hemlock — high-subsonic Mach 0.8 cruise and terminal dive Piranha USV — the test-and-identify method, with its own marine autopilot

Depends on: Edge compute, Flight software.

Retrieved from "subsystems/near-envelope-control.html"