The core component of the project is a ROS package implementing the reinforcement learning process for drone control. The project also includes a Docker image with a prepared simulation of the Bebop Parrot drone in the Gazebo environment under ROS Melodic. Developed package implements a custom OpenAI Gym environment integrated with ROS interfaces for controlling a Bebop Parrot drone in the simulation. Project uses the PPO method from the Stable-Baselines3 library to train a drone control model.

Build the docker image:

./build.sh

Run the container:

./run.sh

Build the ROS package:

cd /root/catkin_ws
catkin_make --pkg bebop_rl

Run Bebop Parrot drone simulator

roslaunch bebop_simulator task1_world.launch 

Run the project package

rosrun bebop_rl rl_train.py

An example of the control policy in action is shown in the GIF below.