Skip to content

Latest commit

 

History

History
83 lines (64 loc) · 2.77 KB

README.md

File metadata and controls

83 lines (64 loc) · 2.77 KB
Raw

Learning Quadrotor Control Using Differentiable Simulation

youtube_video

This repository contains the flightning python package to learn quadrotor control policies using differentiable simulation. If you want to use this code, please cite the following publication (https://arxiv.org/abs/2410.15979):

@misc{heeg2024learningquadrotorcontrolvisual,
      title={Learning Quadrotor Control From Visual Features Using Differentiable Simulation}, 
      author={Johannes Heeg and Yunlong Song and Davide Scaramuzza},
      year={2024},
      url={https://arxiv.org/abs/2410.15979}, 
}

Abstract

This work demonstrates the potential of differentiable simulation for learning quadrotor control. To the best of our knowledge, this work shows the first application of differentiable-simulation-based learning to low-level quadrotor control. We show that training in differentiable simulation significantly outperforms model-free RL in terms of both sample efficiency and training time, allowing a policy to learn to recover a quadrotor in seconds when providing vehicle state and in minutes when relying solely on visual features. The key to our success is two-fold. First, the use of a simple surrogate model for gradient computation greatly accelerates training without sacrificing control performance. Second, combining state representation learning with policy learning enhances convergence speed in tasks where only visual features are observable. These findings highlight the potential of differentiable simulation for real-world robotics and offer a compelling alternative to conventional RL approaches.

Installation

JAX ony supports GPU acceleration on Linux. If you are using Windows or MacOS, you can still run the code, but it will be slower.

Create Environment

mamba create -n flightning python=3.9
mamba activate flightning

Install Flightning

To install the flightning package, run the following commands in the root directory of this repository:

# install flightning requirements
pip install -r requirements.txt
# install flightning
pip install --use-pep517 -e .

Introduction

The examples directory contains jupyter notebooks that demonstrate how to use Flightning to train a neural network to control a quadrotor.

jupyter-lab examples

License

The code in this project is licensed under the GPL-3.0 license.

Portions of this project are based on code from PureJaxRL and gymnax, licensed under the Apache License 2.0. See THIRD_PARTY_LICENSES/APACHE_LICENSE.txt for details.