$$\huge \displaystyle u = K_p e + K_i \int e \, dt + K_d \frac{de}{dt}$$

---

Faithful reimplementations of onboard drone controllers for simulation and research.

Drone Controllers provides array-API implementations of onboard drone controllers for use in simulation and research. It powers Crazyflow's onboard controller simulation and can be used standalone with NumPy, JAX, PyTorch, or any other compliant library.

import numpy as np
from drone_controllers import parametrize
from drone_controllers.mellinger import state2attitude

ctrl = parametrize(state2attitude, "cf2x_L250")

pos  = np.zeros(3)
quat = np.array([0., 0., 0., 1.])  # xyzw, identity (no rotation)
vel  = np.zeros(3)
cmd  = np.zeros(13)  # [x, y, z, vx, vy, vz, ax, ay, az, yaw, rr, pr, yr]

rpyt, int_pos_err = ctrl(pos, quat, vel, cmd)

Documentation

learnsyslab.github.io/drone-controllers — installation, user guide, examples, and API reference.

Features

• Array API standard: works identically with NumPy, JAX, PyTorch, and any compliant library
• Pure functions: no hidden state; integral errors are explicit return values you pass back each step
• JIT compatible: compile any controller with jax.jit or torch.compile without modification
• Batching: add any number of leading batch dimensions; no loops, no special API

Installation

pip install drone-controllers

Developer install with editable mode (pixi required):

git clone https://github.com/learnsyslab/drone-controllers.git
cd drone-controllers
pixi shell

Controllers

Mellinger

A geometric tracking controller based on the Crazyflie firmware, split into three composable stages:

Stage	Function	Input	Output
1	state2attitude	State + 13-element setpoint	RPYT command + position integral error
2	attitude2force_torque	Attitude + RPYT command	Collective force, body torques + angular velocity integral error
3	force_torque2rotor_vel	Force + torques	4 motor speeds [RPM]

Each stage can be used independently or chained into a full pipeline from state setpoint to motor speeds.

Drone models

Pre-fitted parameters ship for the following platforms:

drone_model	Platform
"cf2x_L250"	Crazyflie 2.x, L250 props
"cf2x_P250"	Crazyflie 2.x, P250 props
"cf2x_T350"	Crazyflie 2.x, T350 props
"cf21B_500"	Crazyflie 2.1 Brushless, 500 props

Related packages

Package	Description
crazyflow	GPU-accelerated differentiable drone simulator in JAX. Uses drone-controllers to provide its onboard controller simulation.
drone-models	Drone dynamics models (first-principles and fitted) compatible with NumPy, JAX, and PyTorch.

Testing

pixi run -e tests tests       # unit tests
pixi run -e tests test-docs   # doc snippet tests