A Deep Reinforcement Learning Technique for Autonomous Multirotor Landing on a Moving Platform

Parrot Bebop 2 autonomous landing on a moving platform. The behaviour has been completely learned in simulation without prior human knowledge and by means of deep reinforcement learning techniques. Since the multirotor is controlled in attitude, no high level state estimation is required.

Requirements

• Ubuntu 16.04 LTS
• ROS Kinetic
• OpenCV 3
• Tensorflow
• Boost (c++)
• Rospy
• CvBridge
• Matplotlib
• Numpy

Installation and build

Create workspace and clone respository:

mkdir ~/workspace/drl-landing/src

cd ~/workspace/drl-landing/src

git clone https://github.com/alejodosr/drl-landing

Build ArUco library:

cd ~/workspace/drl-landing/src/drl-landing/code-rl-environment-gazebo/rl_libs/aruco304

mkdir build

cd build

cmake ..

make - j4

Build the workspace:

cd ~/workspace/drl-landing

catkin_make

(Download and install Parrot Bebop 2 Autonomy Driver under ~/workspace/drl-landing/src - https://github.com/AutonomyLab/bebop_autonomy -)

Usage

Coming soon..