[📄 Paper] [🌐 Website] [🤗 Hugging Face]
The Official Implementation of "ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow"
- Clone this repository and create the environment.
git clone https://github.com/Dstate/ODEWorld.git
cd ODEWorld
conda create -n odeworld python=3.10 -y
conda activate odeworld- Install the remaining dependencies.
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt- Download the pretrained checkpoints into
assets/pretrained.
mkdir -p assets/pretrained
for model in \
ODEWorld-PT-Flow-LIBERO \
ODEWorld-PT-Flow-AgiBot \
ODEWorld-Goal-Predictor-LIBERO \
ODEWorld-RAE-LIBERO \
ODEWorld-RAE-AgiBot
do
hf download "ldxxx/${model}" --local-dir "assets/pretrained/${model}"
doneRun the LIBERO examples:
python demo_infer.py --dataset liberoRun the AgiBot examples:
python demo_infer.py --dataset agibotUse --case-ids case_00 to run a single case. Results are written to outputs/<dataset>/<case_id>.
@article{liu-niu2026odeworld,
title={ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow},
author={Liu, Dongxiu and Niu, Haoyi and Cheng, Peng and Gao, Yuan and Kang, Xirui and Teng, Sangli and Sreenath, Koushil and Zhan, Xianyuan},
journal={arXiv preprint arXiv:2607.27924},
year={2026}
}