aerial-kit

Control stack for aerial robots. The core has no ROS or Matplotlib dependency; simulation and visualization are available as optional extras.

For firmware, hardware integration, and the full ROS 2 / simulation stack, visit the repo: github.com/RawFish69/aerial-kit.

What's in it

• aerial_kit.types - SimState, ControlTarget, Capabilities, Wrench, CommandKind, Waypoint
• aerial_kit.interfaces - Controller, DynamicsBackend, Planner ABCs
• aerial_kit.registry - a pluggable component registry (register_airframe/register_controller/etc., create_* factories)
• aerial_kit.airframes - Airframe ABC, MultirotorAirframe (mixer-driven quad/hex/ octo), TwinWingAirframe (elevon + differential-thrust allocation, trim)
• aerial_kit.dynamics - 6-DOF multirotor dynamics, point-mass dynamics, and a hand-rolled 6-DOF fixed-wing model with a flat-plate-blended lift curve, drag polar, and moment derivatives
• aerial_kit.controllers - PID/LQR/MPC position controllers, and FixedWingL1TECSController (L1 lateral guidance + TECS-lite longitudinal control + coordinated-turn attitude PID)
• aerial_kit.guidance - l1_bank_command, tecs_command as standalone functions

Quick start

from aerial_kit.registry import register_builtin_components, create_airframe, create_controller

register_builtin_components()
airframe = create_airframe("quad")
controller = create_controller("pid")
print(airframe.capabilities)

register_builtin_components() here registers only what lives inside aerial_kit itself (airframes, controllers) - it has no ROS or matplotlib dependency and does not know about any host application's own dynamics backends or planners. A host application (like sim_py in the parent repo) registers its own backends/planners into the same registry alongside this.

Simulation and visualization

Install the optional simulator and Matplotlib viewer:

python -m pip install "aerial-kit[sim]"

Run the bundled default quadrotor scenario:

aerial-kit-sim
aerial-kit-sim --example fixed-wing
aerial-kit-sim --no-show --save result.png

Teleop

Launch real-time keyboard teleoperation with one command:

aerial-kit-teleop                          # quadrotor
aerial-kit-teleop --airframe fixed-wing    # twin-wing

Quadrotor — drone-style controls:

Fixed wing — RC-plane-style controls (no rudder, so Q/E biases differential thrust rather than yawing directly):

Shared: X neutralize, P pause, C toggle follow/world camera, -/= zoom, H hide help, Esc exit.

Controls are body-relative, so yawing with Q/E changes where W takes you. Click the plot window first; the HUD shows NO FOCUS when keystrokes are not reaching it, and held keys are released whenever focus is lost. The follow camera is a third-person chase view that stays behind the vehicle and turns with its heading.

The equivalent simulator commands are aerial-kit-sim --teleop and aerial-kit-sim --example fixed-wing --teleop. Without installing the package, run python -m aerial_kit.sim.teleop (add --airframe fixed-wing for the wing) from the repository root, or python examples/quadrotor/teleop.py / python examples/fixed_wing/teleop.py from anywhere (which is also what an IDE Run button does). Teleop needs an interactive Matplotlib backend and fails with an explanatory message if the active backend can only write files.

The public Python API accepts a YAML path, a mapping, or a normalized config:

from aerial_kit.sim import load_config, run_simulation
from aerial_kit.visualization import plot_simulation

config = load_config("examples/quadrotor/config.yaml")
result = run_simulation(config)
print(f"goal error: {result.distance_to_goal:.2f} m")
plot_simulation(result)

Complete configurable examples are included in the repository:

• examples/quadrotor: quad airframe, PID control, and native multirotor dynamics.
• examples/fixed_wing: twin-motor flying wing, Dubins planning, native 6-DOF dynamics, and L1/TECS control.

The fixed-wing example starts at cruise airspeed to represent a hand launch. A fixed wing cannot be initialized at zero velocity like a hovering multirotor.

Status

Early and actively developed. Be clear-eyed about what's actually verified:

• Multirotor (quad/hex/octo airframes, PID/LQR/MPC): verified, working. This is the path flying on real hardware in the parent project.
• Twin-motor wing (TwinWingAirframe, FixedWingL1TECSController, L1/TECS guidance): under development, simulation-only. Not flown on any hardware. Its 6-DOF aero model uses plausible placeholder coefficients, not a fitted model of any specific real airframe, and its guidance gains are validated only in the specific simulated scenarios its own test suite covers - treat it as a research/simulation component, not something to fly as-is.

License

MIT