GRAIL: Generating Humanoid Loco-Manipulation from 3D Assets and Video Priors

GRAIL is a fully digital data-generation pipeline for humanoid loco-manipulation. It composes 3D assets, simulator-ready scenes, robot-proportioned characters, and video foundation model priors to synthesize metric 4D human-object interaction (HOI) trajectories, then retargets them to a Unitree G1 and trains task-general policies for pick-up, whole-body manipulation, sitting, and terrain traversal. Using only GRAIL-generated data, the resulting egocentric visual policies transfer to real-world object pick-up and stair-climbing.

Motion Gallery

Tabletop Pickup	Ground Pickup

Tabletop Manipulation	Ground Manipulation

Sitting	Curb

Slope	Stairs

Sim-to-Real Deployment

Rendered Egocentric Views

Pick-up	Stair-Climbing

Quick Start

Pull the docker image and install local extras (Blender, checkpoints) inside it:

docker pull docker.io/nvgrail/grail:latest

docker run --gpus all -it --shm-size=16g \
    -v /path/to/grail:/workspace/grail \
    docker.io/nvgrail/grail:latest

# inside the container
cd /workspace/grail
bash scripts/setup/install_env_docker.sh   # downloads Blender
bash scripts/setup/download_checkpoints.sh # GEM-SMPL / GEM-SOMA / FoundationPose weights
conda activate grail
source .env                                 # OPENAI_API_KEY, KLING_*, HF_TOKEN

Run any stage end-to-end. Pipeline stages are package entrypoints; invoke them with python -m grail.pipelines.* rather than project-root wrapper scripts.

# 3D asset generation (procedural terrain or AI-generated objects)
python -m grail.pipelines.gen_terrain --type stairs --num 50 --output_dir data/syn_stairs
conda run -n hunyuan python -m grail.pipelines.gen_3d_assets \
    -i configs/gen_3d/example_objects.yaml -o data/gen_example

# 2D HOI generation (Blender + Kling video)
python -m grail.pipelines.gen_2dhoi --dataset ComAsset --category cordless_drill \
    --character kid --results_dir results --video_model_api kling-ai

# 4D HOI reconstruction
python -m grail.pipelines.recon_4dhoi --dataset ComAsset --category cordless_drill --results_dir results

Full install, dataset, and config notes: see Documentation below.

Documentation

Full documentation can be found at docs (rendered HTML) and markdown sources are linked below.

Getting Started

Pipeline

TODOs

Provide quick-start demo script
Release GRAIL manipulation dataset
Release task-general tracking policy checkpoints

Citation

If you find GRAIL useful in your research, please cite:

@misc{grail2026,
  title         = {GRAIL: Generating Humanoid Loco-Manipulation from 3D Assets and Video Priors},
  author        = {Tianyi Xie and Haotian Zhang and Jinhyung Park and Zi Wang and Bowen Wen and Jiefeng Li and Xueting Li and Qingwei Ben and Haoyang Weng and Yufei Ye and David Minor and Tingwu Wang and Chenfanfu Jiang and Sanja Fidler and Jan Kautz and Linxi Fan and Yuke Zhu and Zhengyi Luo and Umar Iqbal and Ye Yuan},
  year          = {2026},
  eprint        = {2606.05160},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  doi           = {10.48550/arXiv.2606.05160},
  url           = {https://arxiv.org/abs/2606.05160},
}

License

This project is released under the NVIDIA License; see LICENSE for details. The Work and any derivative works may be used only non-commercially, except by NVIDIA Corporation and its affiliates. Third-party components are subject to their own licenses.