RoboVoLo is the benchmark task pack for VoLo: A Physical Orchestrator for Open-Vocabulary Long-Horizon Manipulation. It contains 126 RoboLab tasks, 123 USD scenes, benchmark images, and three custom object-asset families.
This repository holds the benchmark tasks and assets. The VoLo orchestrator code — the VLM-driven proxy that runs and evaluates policies on these tasks — lives in the companion repository VoLoAgent.
License overview: Code, tasks, scenes, benchmark images, and NVIDIA-authored assets in this repository are licensed under the Apache License 2.0 unless otherwise noted. Redistributed third-party textures retain their asset-specific CC0-1.0 terms; see ASSET_LICENSES.md and THIRD_PARTY_NOTICES.md. Required external assets are not distributed with this repository and must be obtained separately under their own terms.
- task definitions and metadata in
tasks/,tasks.json, andpack.json; - released benchmark scenes in
scenes/and their images inimages/; - redistributable art-shape, chemistry-cube, and math-cube assets in
assets/objects/; - installers for adding the benchmark to a RoboLab checkout; and
- a downloader for required NVIDIA Isaac SimReady 6.0 assets that are not included in this repository.
The scenes also reference assets shipped by RoboLab v0.3.0. Those RoboLab payloads are not duplicated here.
Clone the required release, then follow RoboLab's environment installation instructions:
git clone --branch v0.3.0 --depth 1 https://github.com/NVlabs/RoboLab.gitRoboLab v0.3.0 natively discovers the robovolo content pack and includes the grouped task predicates required by RoboVoLo. The RoboVoLo installer validates these APIs before changing the RoboLab checkout.
git lfs install
git clone https://github.com/NVlabs/RoboVoLo.git
cd RoboVoLo
git lfs pullInstall the Python dependencies in the same environment you will use to run the download and installation scripts:
python -m pip install -r requirements.txtThis installs the USD Python bindings required to patch the SimReady assets correctly.
Review the external materials notice, the retained NVIDIA Software License Agreement, and the retained Product Specific Terms for NVIDIA AI Products, then run:
python download_simready_assets.py \
--agree-external-terms \
/path/to/RoboLabThe script prints the source, terms, and complete asset list before downloading.
The installer accepts the local RoboLab checkout as its positional argument and installs the custom assets, tasks, and scenes:
python install.py /path/to/RoboLabThe installer requires a fresh installation and stops without changing the checkout if any RoboVoLo-managed destination already exists. Run python install.py --help for usage details.
NOTICE AND DISCLAIMER: This software automatically retrieves, accesses or interacts with external materials. Those retrieved materials are not distributed with this software and are governed solely by separate terms, conditions and licenses. You are solely responsible for finding, reviewing and complying with all applicable terms, conditions, and licenses, and for verifying the security, integrity and suitability of any retrieved materials for your specific use case. This software is provided "AS IS", without warranty of any kind. The author makes no representations or warranties regarding any retrieved materials, and assumes no liability for any losses, damages, liabilities or legal consequences from your use or inability to use this software or any retrieved materials. Use this software and the retrieved materials at your own risk.
The downloader retrieves the 34 required NVIDIA Isaac SimReady 6.0 asset families listed in download_simready_assets.py from NVIDIA's public Omniverse content bucket. These payloads are not present in Git or Git LFS. They are governed by the retained NVIDIA Software License Agreement, the retained Product Specific Terms for NVIDIA AI Products, which incorporate the NVIDIA Omniverse product specific terms, and any applicable asset-specific terms. See the current Omniverse licensing page for the official online source.
tasks/— benchmark task definitionsscenes/— benchmark USD scene compositionsimages/— benchmark scene rendersassets/objects/— included custom object assets and CC0 texturesinstall.py— installs included assets, tasks, and scenes into RoboLabdownload_simready_assets.py— explicitly invoked external-material downloaderrequirements.txt— Python requirements for the download and installation utilitiesLICENSES/— retained third-party and external-material license termsASSET_LICENSES.md— path-level asset license manifestTHIRD_PARTY_NOTICES.md— third-party sources and notices
- VoLoAgent — the VoLo orchestrator: a VLM-driven proxy that runs and evaluates policies on the RoboVoLo benchmark.
- SpaceTools (CVPR 2026) — Tool-Augmented Spatial Reasoning via Double Interactive RL, which equips VLMs with vision and robotic tools for spatial reasoning and real-world manipulation.
Paper: arXiv:2606.07723
Project website: https://chicychen.github.io/VoLo/
@article{chen2026volo,
title = {VoLo: A Physical Orchestrator for Open-Vocabulary Long-Horizon Manipulation},
author = {Chen, Siyi and Hadfield, Hugo and Zook, Alex and Uy, Mikaela Angelina and
Song, Chan Hee and Coumans, Erwin and Yang, Xuning and Ladhak, Faisal and
Qu, Qing and Birchfield, Stan and Tremblay, Jonathan and Blukis, Valts},
journal = {arXiv preprint arXiv:2606.07723},
year = {2026},
eprint = {2606.07723},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.07723}
}External contributions are accepted under the Developer Certificate of Origin. Every contribution must be signed off; see CONTRIBUTING.md.