conda create -n tsmp python=3.7
conda activate tsmp
conda install -c conda-forge git-lfs

In highway-env: python3 setup.py develop
In rl-agents: python3 setup.py develop
In muzero-general: pip install -r requirements.txt

For logging to Weights and Biases wandb login

• All MuZero experiments are run from within muzero-general, our fork of https://github.com/werner-duvaud/muzero-general with all our modifications.
• Set the environment variables RAY_TEMP_DIR and WANDB_USERNAME for Ray and Weights and Biases init.

Each setting defines its own game file under muzero-general/games. All training are run by python muzero_interactive GAME_NAME. Relevant game files are the following (by environment):

• MuZero: highway_env_ttc_flat

• MuZero: roundabout_env_ttc_flat
• Stochastic MuZero: roundabout_env_ttc_flat_stochastic_concat
• Risk-Sensitive MuZero: roundabout_env_ttc_flat_risk_sensitive