Causal-Knowledge-Transfer-Safe-RL

Causal-Based Knowledge Transfer in Safe Reinforcement Learning

Setup

1. Clone this repository.
2. Install Sumo https://sumo.dlr.de/docs/Installing/index.html
   • Remember to set your SUMO_HOME environment variable https://sumo.dlr.de/docs/Basics/Basic_Computer_Skills.html#sumo_home
   • Current Version: 1.20.0
3. Create a Conda Environment conda create --name CKT-4-SafeRL python=3.11
   • Stable-Baselines + Tensorflow do not support python 3.12 yet
4. Install Sumo Python Tools pip install -r $SUMO_HOME/tools/requirements.txt
5. Install SumoRL https://lucasalegre.github.io/sumo-rl/install/install/
   • Use pip --version to check whether the correct environment is used
   • the path should look similar to this: /Users/username/miniconda3/envs/CKT-4-SafeRL/lib/python3.11/site-packages/pip
   • pip install sumo-rl
6. Install PyTorch >= 1.13 https://pytorch.org/get-started/locally/
7. Install Stable-Baselines3 https://stable-baselines3.readthedocs.io/en/master/guide/install.html
   • Use pip install setuptools==65.5.0
   • Use pip install 'stable-baselines3[extra]' (' required for some shells)
8. Install Tensorflow https://www.tensorflow.org/install
   • Use pip install tensorflow
     • There might be a dependency issue: ortools 9.10.4067 requires protobuf>=5.26.1, but you have protobuf 4.25.4 which is incompatible.
     • pip install ortools==9.9.3963 fixes that
9. Install Jupyter Notebook https://jupyter.org/install
   • Use pip install notebook
10. Install seaborn https://seaborn.pydta.org/installing.html
    • Use pip install seaborn
11. Install gCastle
    • Use pip install gcastle==1.0.4rc1
12. Install tqdm
    • Use pip install tqdm
    • pip install ipywidgets