Robust inverse reinforcement learning through Bayesian theory of mind

Bayesian MAP simultaneous estimation of reward and dynamics for offline model-based inverse reinforcement learning. Pytorch implementation of paper.

Usage

Environment set up:

conda env create -f environment.yml
conda activate irl

D4RL MuJoCo experiments

Optional dataset download:

python scripts/download_d4rl_data.py

Run BTOM IRL:

sh scripts/irl/train_btom.sh

Gridworld experiments

Create demonstrations and run IRL (two-stage IRL as default, change settings in the .sh script):

sh scripts/tabular/create_gridworld_demonstrations.sh
sh scripts/tabular/train_gridworld.sh

Implemented algorithms

See src/agents and src/algo for additional implemented RL and IRL algorithms.