Flight models

Three airframe families, one set of conventions, one base class. They compose because they agree on frames — see Frames and conventions.

multirotor

Quadrotor

6DOF Newton-Euler rigid body with per-motor first-order dynamics and a configurable mixer.

N rotors

Multirotor

Any even rotor count and any arm geometry, with the mixing matrix derived from rotor positions and spin directions rather than hard-coded.

aerodynamic

Fixed-wing

Full Beard & McLain coefficient build-up: stability derivatives, induced drag, post-stall lift, momentum-theory propeller.

hybrid

VTOL tilt-rotor

Rotors that tilt from hover to cruise, sharing the fixed-wing airframe model, with a rate-limited tilt actuator.

What each one models

	Quadrotor	Fixed-wing	Tilt-rotor
Rigid-body 6DOF	yes	yes	yes
Integrator	RK4	RK4	RK4
Actuator dynamics	per-motor first order	surface rate limits	tilt slew limit
Aerodynamics	linear body drag	full coefficient build-up	shared wing model
Stall	—	yes, flat-plate blend	yes, shared
Propulsion	thrust-coefficient motors	momentum-theory propeller	tilting rotor thrust
Trim solver	hover wrench (closed form)	numerical, compute_trim	—
Presets	6 (4 quad, hex, coaxial octo)	4	1
State velocity	world ENU	body FLU	world ENU
Control input	[T, τx, τy, τz]	[δe, δa, δr, δt]	[T, τx, τy, τz, tilt]

What they deliberately do not model

Being explicit about this matters more than the feature list. None of the models include:

• Wind, gusts or turbulence. Airspeed equals ground speed everywhere. Adding a wind field is a natural extension; the aerodynamics already work from an air-relative velocity, so it would enter in one place.
• Ground effect, rotor-wake interaction, or blade flapping.
• Compressibility or Reynolds-number effects. Coefficients are constant.
• Structural flexibility. Every airframe is rigid.
• Actuator failure or saturation dynamics beyond simple rate and range limits.
• Sensor mounting on the dynamics. Sensors read the true state and add their own noise models; they do not load the airframe.

The models are built to be correct within their stated scope and readable, not to be a certified flight-dynamics package. See the safety note.

The shared base

Every airframe subclasses UAVBase, which owns the state vector, RK4 integration and reset semantics:

from uav_sim.vehicles.base import UAVBase, UAVParams

class MyAirframe(UAVBase):
    @property
    def state_dim(self) -> int: ...

    @property
    def control_dim(self) -> int: ...

    def _dynamics(self, state, control):
        """Return dx/dt."""

Quadrotor predates the base class and keeps its own step() so that motor states integrate alongside the rigid body; it follows the same state layout and conventions.

Choosing one

• Learning control theory, or anything indoors → quadrotor. It hovers, which makes every experiment cheap to set up.
• Range, endurance, or aerodynamics → fixed-wing. It cannot hover, so it forces you to think about trim, stall and coordinated turns.
• Transition control → tilt-rotor. The interesting part is the handover of lift from rotors to wing, which is a genuinely hard control-allocation problem.

Common pitfalls

Positive pitch is nose-down

A consequence of the Forward-Left-Up body frame. Use aircraft.pitch_up when displaying an attitude to a human.

A fixed wing needs a big world

The default Aerosonde cruises at 35 m/s and stalls near 13 m/s. It cannot fly in the 30 m worlds the quadrotor simulations use — it will leave the volume in under a second. Use MINI_TRAINER for small worlds, or scale the world up.

Quadrotor motors start stopped

reset() zeroes the motor speeds. Commanding hover thrust to stopped motors still drops the aircraft while they spin up. Pre-load them with motor.reset(motor.thrust_to_omega(hover / 4)) when you want an undisturbed hover.