Use these repositories to train a drone control policy in simulation and deploy it to a real CrazyFlie 2.1 drone.
Contents:
- phoenix-drone-simulation Contains a PyBullet based simulator and a PPO implementation in PyTorch which can be used with recurrent neural networks
- crazyflie-fw-neural-ctrl Source code for the execution of neural network based controllers on the CrazyFlie's microcontroller
- crazyflie-clients-python Tools for flashing the binary onto the microcontroller and uploading neural network weights to the drone
- Thesis Daniel Stuemke.pdf My master's thesis, which describes the domain randomization scheme and the reinforcement learning algorithm