Causal Policy Gradient for Whole-Body Mobile Manipulation

Jiaheng Hu, Peter Stone, Roberto Martin-Martin

RSS2023

Setup

1. Clone this repo and its submodules:

git clone https://github.com/JiahengHu/CausalMoMa.git --recursive

2. Install the cloned iGibson-CausalMoMa, Minigrid-CausalMoMa and sb3-CausalMoMa, following the respective README.md instructions.

3. Download the required iGibson data. Download HSR mesh data and extract it into iGibson-CausalMoMa/igibson/data/assets/models/hsr

Causal Inference

1. Download pre-collected Causal inference data and put them into data/. Alternatively, collect new data by running:

# iGibson data
python collect_igibson_data.py

# Minigrid data
python collect_igibson_data.py

2. Run causal discovery with one of the config file provided:

python causal_inference.py --config PATH_TO_CONFIG

# e.g., for minigrid
python causal_inference.py --config configs/minigrid_full.json

Results will be stored inside causal/.

Policy Learning

The inferred causal matrix is already put inside train.py

# HSR with Causal MoMa
python train.py -sc --robot hsr

# HSR with Vanilla PPO
python train.py -fc --robot hsr

# Fetch with Causal MoMa
python train.py -sc --robot fetch

# Fetch with Vanilla PPO
python train.py -fc --robot fetch

Results will be stored inside log_dir/.

Citing

@inproceedings{hu2023causal,
  title={Causal Policy Gradient for Whole-Body Mobile Manipulation},
  author={Hu, Jiaheng and Stone, Peter and Mart{\'\i}n-Mart{\'\i}n, Roberto},
  booktitle={arXiv preprint arXiv:2305.04866},
  year={2023}
}