Code for the NeurIPS 2021 paper "Safe Reinforcement Learning by Imagining the Near Future" by Garrett Thomas, Yuping Luo, and Tengyu Ma.
Some code is borrowed from Force.
We are using Python 3.8. The required packages can be installed via
pip install -r requirements.txt
You also must set the ROOT_DIR in code/defaults.py.
This is where experiments' logs and checkpoints will be placed.
Once setup is complete, run the code using the following command:
python main.py -c config/ENV.json
where ENV is replaced appropriately. To override a specific hyperparameter, add -s PARAM VALUE where PARAM is a string.
Use . to specify hierarchical structure in the config, e.g. -s alg_cfg.horizon 10.