This repository contains the code for the paper Safe Reinforcement Learning Using Black-Box Reachability Analysis. Demo video

The repository provides two ros packages: brsl and brsl_msgs. To get everything up and running, we need to install ros and some other dependencies. Everything is tested on ubuntu 20.04 LTS

you can find the installation steps for ros in the official link.

You need to install the following libraries.

python -m pip install absl-py
python -m pip install gin
python -m pip install tensorflow
python -m pip install gin-config
python -m pip install tf-agents
python -m pip install scipy
python -m pip install joblib
python -m pip install tensorboardX
python -m pip install torch
python -m pip install omegaconf==2.0
python -m pip install hydra-core==1.1.0
python -m pip install mbrl==v0.1.4

If you want to utilize gpu for your training, you should install tf-gpu and torch cuda instead.

To run the agents, first, make a catkin_ws.

cd
mkdir -p catkin_ws/src
cd catkin_ws
catkin_make

Then copy both brsl and brsl_msgs to catkin_ws/src in your home directory.

Finally, run:

cd ~/catkin_ws/src
catkin_make

This should build the repository.

To run the scripts, go to your catkin_ws directory, source the workspace, and run both the environment and the launch script for the agent.

roscore

Then, in a second terminal run the environment:

cd ~/catkin_ws
source devel/setup.bash
rosrun brsl Turtlebot_environment

Finally, open a third terminal and run the agent

cd ~/catkin_ws
source devel/setup.bash
roslaunch brsl turtlebot.launch