Real-World Attitude Rate Crazyflie 2.1 Controller

Easop Lee, Samuel A. Moore, and Boyuan Chen
Duke University

Offboard attitude rate control on Crazyflie 2.1 w/ vicon mocap data.

---

⚠️ Important Notice
This repository must be used together with the Sym2Real framework. Please check out this repository first.

• The Sym2Real repo provides modular dynamics models, controllers, and simulation environments.
• This Crazyflie controller repo provides the hardware interface for the quadrotor real-world experiments.

Please keep both repositories in sync when running experiments. Updates or mismatches between the two may cause unexpected errors.

---

🔗 Syncing with Sym2Real

Make sure you clone the Sym2Real repo, and follow its installation instructions first.

Then, clone this repository next to the Sym2Real (e.g. if the parent folder is Desktop):

Desktop/
├── sym2real
└── crazyflie-controller

git clone [email protected]:easoplee/crazyflie-controller.git

🔧 Building and flashing firmware

Flash the firmware to the Crazyflie hardware.

cd crazyflie-firmware
git submodule init
git submodule update --recursive

make cf2_defconfig
make

cfloader flash build/cf2.bin stm32-fw -w radio://0/70/2M/E7E7E7E701 # change the radio address accordingly

🛠️ Installation

Install additional packages to the previously created sym2real conda environment (Sym2Real).

conda activate sym2real

pip install cfclient==2024.7.1

cd crazyflie-lib-python
pip install -e .

✅ Sanity Check

First, perform this sanity check test to make sure all the components are set up correctly. Make sure you have enough space since crazylfie willl start moving.

cd tests
python simple_sanity_check.py  # to check cflib: expected to move its propellers 

Then, start the vicon streaming. We assume the name is "cf1" here.

self.vicon = ViconReader(topic="/vicon/cf1/cf1")

The folowing script should print all the states (both from vicon and cflib) and send commands to the crazyflie. Crazyflie should move its propellers without flying away.

python readsend_sanity_check.py # check reading and sending commands, tests the whole pipeline

🚀 Quickstart

Make sure you sync the saved model folders from the experiments in Sym2Real. Change the parent folder in setup.sh. We assume the following structure:

Desktop/
├── sym2real
└── crazyflie_controller

cd crazyflie_controller
source setup.sh

▶️ Run Base Model - Zero-shot (e.g. Symbolic Regression on Crazyflie)

python main.py \
    dynamics_model="symbolic_regression" \
    model_path="{path_to_sr_base_model_experiment}"

e.g. model_path="/home/generalroboticslab/Desktop/sym2real/example_results/CF2HoverMujocoEnvAttitudeRate-v0/hover/env_params_wind_x_0_wind_y_0_wind_z_0_mass_0.027_com_x_0.00/seed_0/symbolic_regression/2025.07.13:142911"

These prompts will show up to ask if you want to change the name of the folder or whether you want to collect multiple trajectories.

1. Do you want to create a new set of experiments? Then quit and fix the config file. [y/n]: 

# type "n" unless you want to change the saved folder path into config file

2. Please enter the trial number:

# type the trial number (e.g. for zero-shot, type 1. After a model update, for re-deployment type 2, ...)

3. Collect another trajectory? [y/n]:

# if you want to collect another trajectory with this model, hit "y".

4. Enter traj number (reset drone before entering this)

# which trajectory is this. Type an integer.

Then, to train the residual model on the collected real world data:

python train_model.py \
    dynamics_model="residual_mlp" \
    dynamics_model.model_a_path="{path_to_sr_base_model_experiment}"

e.g. dynamics_model.model_a_path="/home/generalroboticslab/Desktop/sym2real/example_results/CF2HoverMujocoEnvAttitudeRate-v0/hover/env_params_wind_x_0_wind_y_0_wind_z_0_mass_0.027_com_x_0.00/seed_0/symbolic_regression/2025.07.13:142911"

🪄 Run Residual Model - Finetuned (e.g. Symbolic Regression + Residual MLP on Crazyflie)

python main.py \
    dynamics_model="residual_mlp" \
    dynamics_model.model_a="symbolic_regression" \
    dynamics_model.model_a_path="{path_to_sr_base_model_experiment}" \
    model_path="{path_to_residual_mlp_model_experiment}""

# one folder above the saved .pkl path
e.g. model_path="/home/generalroboticslab/Desktop/crazyflie_controller/real_crazyflie_example_results/2pennies/CF2HoverMujocoEnvAttitudeRate-v0/seed_0/symbolic_regression_residual_mlp/dynamics_model/mlp_model_w_dataset_size_400"