Elastic Decision Transformer

[Project Page] [Paper]

Elastic Decision Transformer, Yueh-Hua Wu, Xiaolong Wang, Masashi Hamaya, NeurIPS 2023.

Elastic Decision Transformer is a novel Decision Transformer approach that enables trajectory stitching by adopting different history length.

Bibtex

@article{wu2023elastic,
  title={Elastic Decision Transformer},
  author={Wu, Yueh-Hua and Wang, Xiaolong and Hamaya, Masashi},
  journal={arXiv preprint arXiv:2307.02484},
  year={2023}
}

Installation

We prepared a Dockerfile and bash scripts to set up the environment.

1. Build the Docker image and start a Docker container

# Download the code from this repo
git clone https://github.com/kristery/Elastic-DT.git
cd Elastic-DT
bash build_image.sh
bash start_container.sh

Training

1. Download D4RL datasets

cd /workspace
python data/download_d4rl_datasets.py

2. Train the EDT agent

python scripts/train_edt.py --env hopper --dataset medium-replay

Evaluation

python scripts/eval_edt.py --chk_pt_name saved_model_name_from_training.pt

Acknowledgement

The implementation of EDT is based on min-decision-transformer