Deep Drone Acrobatics

This repo contains the code associated to our paper Deep Drone Acrobatics.

Citing

If you use this code in an academic context, please cite the following publication:

RSS 2020 Video Pitch: Youtube

@inproceedings{kaufmann2020RSS,
  title={Deep Drone Acrobatics},
  author={Kaufmann, Elia and Loquercio, Antonio and Ranftl, Ren{\'e} and M{\"u}ller, Matthias and Koltun, Vladlen and Scaramuzza, Davide},
  booktitle={Proceedings of Robotics: Science and Systems}, 
  year={2020}, 
  address={Corvalis, Oregon, USA}, 
  month={July}, 
  doi={10.15607/RSS.2020.XVI.040} 
}

Installation

Requirements

The code was tested with Ubuntu 18.04, ROS Melodic, Anaconda v4.8.3. Different OS and ROS versions are possible but not supported.

Step-by-Step Procedure

Use the following commands to create a new catkin workspace and a virtual environment with all the required dependencies.

export ROS_VERSION=melodic
mkdir drone_acrobatics_ws
cd drone_acrobatics_ws
export CATKIN_WS=./catkin_dda
mkdir -p $CATKIN_WS/src
cd $CATKIN_WS
catkin init
catkin config --extend /opt/ros/$ROS_VERSION
catkin config --merge-devel
catkin config --cmake-args -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS=-fdiagnostics-color
cd src

git clone https://github.com/uzh-rpg/deep_drone_acrobatics.git
vcs-import < deep_drone_acrobatics/dependencies.yaml

#install extra dependencies (might need more depending on your OS)
sudo apt-get install libqglviewer-dev-qt5

Before continuing, make sure that your protobuf compiler version is 3.0.0. To check this out, type in a terminal protoc --version. If This is not the case, then check out this guide on how to do it.

Note that building the VINS-Mono package requires to install the Ceres Solver. You can find installation instructions here or here.

# Build and re-source the workspace
catkin build
. ../devel/setup.bash

# Create your learning environment
cd deep_drone_acrobatics
conda create --name drone_flow python=3.6
conda activate drone_flow
# Install (in an hacky way) python requirements
pip install -r pip_requirements.txt
conda install --file conda_requirements.txt

Let's do a Power Loop

Once you have installed the dependencies, you will be able to fly in simulation with our pre-trained checkpoint. You don't need necessarely need a GPU for execution. Note that if the network can't run at least at 15Hz, you won't be able to fly successfully.

Lauch a simulation! Open a terminal and type:

cd drone_acrobatics_ws
source catkin_dda/devel/setup.bash
roslaunch fpv_aggressive_trajectories simulation.launch

Run the Network in an other terminal:

cd
cd drone_acrobatics_ws
. ./catkin_dda/devel/setup.bash
conda activate drone_flow
python test_trajectories.py --settings_file=config/test_settings.yaml

Train your own acrobatic controller

You can use the following commands to generate data in simulation and train your model on it. The trained checkpoint can then be used to control a physical platform (if you have one!).

Generate data

Launch the simulation in one terminal

cd drone_acrobatics_ws
source catkin_dda/devel/setup.bash
roslaunch fpv_aggressive_trajectories simulation.launch

Launch data collection (with dagger) in an other terminal

cd
cd drone_acrobatics_ws
. ./catkin_dda/devel/setup.bash
conda activate drone_flow
python iterative_learning_trajectories.py --settings_file=config/dagger_settings.yaml

It is possible to change parameters (number of rollouts, history length, etc. ) in the file dagger_settings.yaml. Keep in mind that if you change the network (for example by changing the history length), you will need to adapt the file test_settings.yaml for compatibility.

Train the Network

If you want to train the network on data already collected (for example to do some parameter tuning) use the following commands. Make sure to adapt the settings file to your configuration.

cd
cd drone_acrobatics_ws
. ./catkin_dda/devel/setup.bash
conda activate drone_flow
python train.py --settings_file=config/train_settings.yaml

Test the Network

To test the network you trained, adapt the test_settings.yaml with the new checkpoint path. Then follow the instruction to test!