Add dataset card

README.md

@@ -0,0 +1,213 @@
+---
+license: other
+license_name: nvidia-open-model-license
+license_link: >-
+  https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/
+tags:
+  - humanoid-locomanipulation
+  - whole-body-control
+  - human-object-interaction
+  - video-to-motion
+  - reinforcement-learning
+  - physics-simulation
+  - isaac-sim
+  - unitree-g1
+  - smpl-x
+  - 4d-hoi-reconstruction
+library_name: GRAIL
+---
+
+# Dataset Overview
+
+###  Description:
+
+**GRAIL** (**G**enerating Humanoid Loco-Manipulation from Video **F**oundation Models) is a dataset of physics-validated 4D human-object interaction (HOI) trajectories retargeted to the **Unitree G1** humanoid robot. Each motion is the output of an end-to-end pipeline that (1) generates a synthetic interaction video from a 3D asset, (2) reconstructs the underlying 4D HOI (SMPL-X human pose + object 6-DoF) from that video, (3) retargets the human motion to the G1 skeleton, and (4) validates the trajectory in physics simulation by training a reinforcement-learning (RL) tracker against it — the released motion data is the output of the RL tracking policy.
+
+The release is partitioned by HOI category. Each motion ships with: the source synthetic video, the 4D HOI reconstruction (SMPL-X + object pose), the retargeted G1 robot trajectory, the post-RL object trajectory, and the object's USD asset (textures preserved). This layout is byte-compatible with the GRAIL training stack, so the release can be dropped into a `motion_lib_cfg.motion_file` / `object_motion_file` / `object_usd_path` slot without any conversion.
+
+The repo also ships the **submodule checkpoints** required to re-run the full GRAIL pipeline end-to-end (GEM-SMPL human pose estimation + FoundationPose object 6-DoF tracking).
+
+##  License/Terms of Use:
+
+Use of the source code is governed by the Apache License, Version 2.0. Use of the associated dataset and checkpoint weights is governed by the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/).
+
+##  Deployment Geography:
+
+Global
+
+##  Use Case:
+
+GRAIL is intended for use by individuals and professionals in fields such as robotics learning, machine learning, computer vision, and physics-based animation. Specific use cases include:
+
+* **4D HOI reconstruction** — study 4D human-object interaction from a paired dataset of (synthetic video, SMPL-X recon, object 6-DoF, retargeted humanoid trajectory).
+* **Humanoid policy training** — supervise RL or imitation-learning trackers on physically validated reference motions to learn whole-body loco-manipulation skills on the Unitree G1.
+* **Sim-to-real transfer** — use the G1 trajectories directly as targets for a deployable controller, or as kinematic references for a learned residual policy.
+
+##  Release Date:
+
+Hugging Face: 06/03/2026
+
+##  Reference(s):
+
+* **GRAIL**: Generating Humanoid Loco-Manipulation from Video Foundation Models — The GRAIL Authors, 2026
+* **GENMO**: Generalist Model for Human Motion — Li et al., ICCV 2025 (used for SMPL-X / SOMA human pose estimation; `nvidia/GEM-X`)
+* **FoundationPose**: Wen et al., CVPR 2024 (used for object 6-DoF pose tracking)
+* **SMPL-X**: Pavlakos et al., CVPR 2019
+* Project page: <https://research-staging.nvidia.com/labs/dair/grail/>
+* Paper: <https://research-staging.nvidia.com/labs/dair/grail/static/pdf/grail_paper.pdf>
+* Code: <https://github.com/NVlabs/GRAIL>
+* Documentation: <https://nvlabs.github.io/GRAIL/>
+
+##  Dataset Layout:
+
+```
+nvidia/PhysicalAI-Robotics-Locomanipulation-GRAIL/
+├── data/<hoi_category>/
+│   ├── recon/<hoi_category>__<object>__<NNN>.pkl       # 4D HOI recon (SMPL-X + object 6-DoF, world frame)
+│   ├── video/<hoi_category>__<object>__<NNN>.mp4       # source synthetic video
+│   ├── robot/<hoi_category>__<object>__<NNN>.pkl       # G1 robot trajectory (29 body DOFs + hand_dof_pos)
+│   ├── objects/<hoi_category>__<object>__<NNN>.pkl     # post-RL object 6-DoF trajectory
+│   ├── meta/<hoi_category>__<object>__<NNN>.pkl        # per-motion metadata (lengths, contact flags, source IDs)
+│   ├── object_usd/<hoi_category>__<object>__<NNN>.usd  # OpenUSD object asset
+│   │   └── textures/<basename>/                          # per-USD texture subdir, refs rewritten in the USD
+└── checkpoint/
+    ├── GEM-SMPL/                                         # SMPL-X human pose estimation weights (HMR2, ViTPose, VIMO, YOLO, HMR4D)
+    └── FoundationPose/weights/                           # object 6-DoF estimator (refiner + scorer)
+```
+
+The 3-digit `NNN` index restarts at 0 within each `<object>`. Filenames are deterministic so that a re-export of the same source data produces the same on-disk layout.
+
+###  Dataset Statistics per HOI Category:
+
+| Category | What it is | 3D asset source | # objects | # motions | Seq. length | Total frames |
+|---|---|---|---:|---:|---:|---:|
+| `pickup_table` | Tabletop pick-up — grasp an object from a table surface, lift, transport | RoboCasa-derived 3D assets | 685 | 2,991 | 10 s | 747,750 |
+| `pickup_ground` | Ground pick-up — kneel to grasp an object from the floor and stand back up | RoboCasa-derived 3D assets | 631 | 1,613 | 15 s | 611,625 |
+
+Additional categories (tabletop / ground manipulation, sitting, terrain locomotion — curb / slope / stairs) are planned for subsequent releases.
+
+##  Additional Statistics:
+
+| Field | Value |
+|---|---|
+| Synthetic video fps | 24 fps |
+| Reconstructed motion fps | 24 fps |
+| Released trajectory rate | 25 Hz |
+| Reconstructed body model | SMPL-X (75 body DOFs + 45 × 2 hand DOFs) |
+| Robot platform | Unitree G1 (29 body DOFs + 7 × 2 hand DOFs) |
+| Modalities per motion | Video (mp4), 4D HOI recon (pkl), robot traj (pkl), object traj (pkl), meta (pkl), object asset (USD + textures) |
+
+
+##  Data Collection Method:
+
+Hybrid — Automatic. Each motion is the deterministic output of the GRAIL pipeline:
+
+1. **3D asset acquisition** — RoboCasa-derived meshes, AI-generated meshes from Hunyuan3D-2.1, or procedural terrain assets. No real-world scans of identifiable objects.
+2. **2D HOI synthesis** — a Blender simulation places a SMPL-X-rigged character interacting with the object; the rendered video is passed through Kling AI to add photorealistic appearance.
+3. **4D HOI reconstruction** — SMPL-X body pose recovered via GEM-SMPL (HMR2 + ViTPose + VIMO + HMR4D); object 6-DoF via FoundationPose conditioned on a SAM2 mask and a MoGe depth prior; jointly optimized in a multi-stage HOI optimizer.
+4. **Retargeting** — SMPL-X human pose is retargeted to the Unitree G1 skeleton via the [GMR](https://github.com/YanjieZe/GMR) IK + temporal-smoothing engine. Hand DOFs and per-motion USD assets are assembled in the same pass.
+5. **RL tracking** — the retargeted motion is used as a tracking reference for a SONIC policy in Isaac Lab. The post-RL object trajectory is the one realized by the simulated G1 + object under contact dynamics — guaranteed to be physically feasible by construction.
+
+##  Labeling Method:
+
+Automated. All annotations (SMPL-X body parameters, object 6-DoF, contact flags, per-frame joint positions) are produced by the GRAIL reconstruction stack; no manual labeling.
+
+##  Visualization:
+
+Released data can be rendered into kinematic-replay MP4s using [GRAIL data visualization](https://nvlabs.github.io/GRAIL/visualization.html). The output can then be browsed using [GRAIL web visualizer](https://nvlabs.github.io/GRAIL/web_visualizer.html) for hover-to-play previews.
+
+```bash
+git clone https://github.com/NVlabs/GRAIL.git && cd GRAIL
+# follow the installation guide to set up the `sonic` conda env
+conda activate sonic
+export DISPLAY=:1
+
+# Batch — render up to 16 motions (default), defaults match the release (xyzw quats)
+bash grail/visualization/scripts/visualize.sh \
+    data/pickup_table
+
+# Or render a single motion from its robot/<seq>.pkl
+bash grail/visualization/scripts/visualize_single.sh \
+    data/pickup_table/robot/pickup_table__apple_0__000.pkl
+```
+
+Outputs land alongside the motion library so the same directory works as input to the web visualizer:
+
+```
+data/<hoi_category>/vis/
+├── <seq>.mp4                    one per motion
+├── all_motions_combined.mp4     concat (only when --max_videos = 0)
+└── examples_grid.mp4            4×4 or 2×2 grid (only when --max_videos = 0)
+```
+
+Knobs you may want to set:
+
+| Arg / env | Default | Notes |
+|---|---|---|
+| `max_videos` (2nd positional) | `16` | Cap on motions rendered. Pass `0` to render the whole library and produce the concat / grid MP4s. |
+| `cam_offset` (3rd positional) | `1.5,-1.5,1.0` | Camera position relative to the motion centroid. |
+
+##  Disclaimer
+
+NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this dataset and the downstream models trained on it meet requirements for the relevant industry and use case and addresses unforeseen product misuse.
+
+##  Ethical Considerations:
+
+GRAIL trajectories are synthetic. No real individuals appear in the source videos, SMPL-X reconstructions, or any other modality — the entire pipeline is synthetic-character-only (the body model is parametric SMPL-X driven by retargeted character animation; no real-person mocap appears in the released motions). The 3D objects are AI-generated, procedural, or licensed from synthetic asset libraries (e.g. RoboCasa).
+
+Users training policies on GRAIL are responsible for the safety properties of those policies once deployed on physical humanoids; the dataset itself is a kinematic reference and does not encode safety constraints, controller stability margins, or hardware torque/velocity envelopes. The Unitree G1 trajectories are guaranteed physically feasible **in the Isaac Lab simulation environment under the SONIC tracker** — sim-to-real transfer requires additional validation by the integrating team.
+
+For more detailed information on ethical considerations for the upstream models GRAIL builds on, see the corresponding model cards: [`nvidia/GEM-X`](https://huggingface.co/nvidia/GEM-X) (human pose estimation) and the FoundationPose project page.
+
+##  Bias
+
+| Field | Response |
+|---|---|
+| Participation considerations from adversely impacted groups in dataset design and testing: | The dataset contains no real individuals. The synthetic 3D characters used to render source videos are drawn from a finite library; their demographic diversity — skin tone, body proportions, age group, sex — is bounded by that library and may not proportionally represent all global populations. Body proportions and appearances common in East Asian, South Asian, Sub-Saharan African, and other underrepresented groups may be less prevalent in the synthetic distribution. The downstream G1 retargeting collapses character-specific body shape onto a single G1 skeleton, so identity-level appearance variation does not propagate into the released robot trajectories. No adversely impacted groups were formally consulted during dataset design or testing. |
+| Measures taken to mitigate against unwanted bias: | (1) **Identity-agnostic output**: The released robot trajectories are encoded in the G1 joint space, which is appearance- and identity-free by construction; (2) **Motion diversity decoupled from appearance**: each character drives a wide variety of motions sampled across HOI categories; (3) **No real-video sampling bias**: the source videos are fully synthetic, so the dataset does not inherit demographic biases from internet-scraped video corpora. |
+| Bias Metric (If Measured): | No formal demographic bias metric has been measured for the dataset. |
+
+##  Explainability
+
+| Field | Response |
+|---|---|
+| Intended Task/Domain: | Humanoid whole-body loco-manipulation — supervising RL or imitation-learning controllers on physically validated reference motions for the Unitree G1. |
+| Dataset Type: | Trajectory dataset (per-motion robot + object 6-DoF + source video + SMPL-X recon). |
+| Intended Users: | Robotics learning researchers; machine-learning engineers; humanoid control researchers; computer-vision and graphics researchers working on HOI. |
+| Output: | Per-motion: G1 robot trajectory `(T, 29)` + hand DOFs, object 6-DoF `(T, 7)` (xyz + quat), input video (mp4), 4D HOI recon (SMPL-X parameters + object pose, world frame), per-motion metadata (`meta/*.pkl`), object asset (`*.usd` + textures). |
+| Describe how the dataset was produced: | A six-stage automated pipeline. (1) 3D asset acquisition — RoboCasa / Hunyuan3D / procedural; (2) physics-simulated character-object interaction in Blender → frame-by-frame render; (3) Kling-AI re-render for photorealistic appearance; (4) 4D HOI reconstruction — SMPL-X via GEM-SMPL, object pose via FoundationPose; joint multi-stage optimizer; (5) retargeting to Unitree G1 via GMR; (6) RL validation in Isaac Lab using SONIC. |
+| Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of: | Not formally tested across demographic subgroups. Because the released trajectories are in G1 joint space rather than per-character body space, per-group outcome variation does not propagate to the dataset's primary downstream use (controller training). |
+| Technical Limitations & Mitigation: | (1) **Single robot platform** — G1 only; cross-embodiment retargeting requires additional work. (2) **Synthetic-to-real domain gap** — source videos are synthetic, so visual-feature-based downstream use (e.g. vision-conditioned policies) may need real-video fine-tuning. (3) **No tactile / force annotations** — the dataset is kinematic + 6-DoF only; contact forces are not exposed. | 
+| Verified to have met prescribed NVIDIA quality standards: | Yes. |
+| Performance Metrics: | Per-motion physical feasibility is verified by construction (released motions are by definition those a trained SONIC tracker can follow). |
+| Potential Known Risks: | (1) **Sim-to-real assumption mismatch** — trajectories that succeed in Isaac Lab may not succeed on a physical G1 without additional residual learning, system-identification, or torque-limit checks. (2) **Object-asset license inheritance** — released USD assets inherit the license of their upstream source (RoboCasa / Hunyuan3D); downstream users should confirm any application-specific redistribution constraints. |
+| Licensing: | Use of the source code is governed by the [Apache License, Version 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md). Use of the associated dataset is governed by the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). |
+
+##  Privacy
+
+| Field | Response |
+|---|---|
+| Generatable or reverse engineerable personal data? | No. The dataset contains no real-person data at any stage — input videos are synthetic renders of synthetic characters; reconstructions are SMPL-X parameters of those synthetic characters; the released robot trajectories are in Unitree G1 joint space and encode no identity, appearance, or biometric signal. |
+| Personal data used to create this dataset? | No. The upstream GEM-SMPL human-pose checkpoints were trained on internally generated synthetic video data (see the [`nvidia/GEM-X`](https://huggingface.co/nvidia/GEM-X) model card). Motion capture animations used during character rigging upstream are recorded from real performers but were retargeted to fixed skeletons that strip identity-specific biometric signals. |
+| Was consent obtained for any personal data used? | Yes (upstream — see `nvidia/GEM-X` for the consent chain on upstream mocap). |
+| Description of methods implemented in data acquisition or processing, if any, to address the prevalence of personal data in the training data: | The release pipeline is end-to-end synthetic. The source character meshes (RenderPeople / NVIDIA Digital Humans / RoboCasa-derived avatars) are synthetic. Re-renders via Kling AI are conditioned on synthetic frames, not real footage. No real-world camera capture is performed at any pipeline stage. |
+| How often is dataset reviewed? | Datasets are initially reviewed upon addition; subsequent reviews are conducted as needed or upon request for changes. |
+| Is a mechanism in place to honor data subject right of access or deletion of personal data? | Not Applicable — the dataset contains no real-person data. |
+| If personal data was collected for the development of the dataset, was it collected directly by NVIDIA? | No |
+| If personal data was collected for the development of the dataset by NVIDIA, do you maintain or have access to disclosures made to data subjects? | Not Applicable |
+| If personal data was collected for the development of this dataset, was it minimized to only what was required? | Yes. |
+| Was data from user interactions with the dataset (e.g. user input and prompts) used to extend the dataset? | No |
+| Does data labeling (annotation, metadata) comply with privacy laws? | Yes — all annotations are automatically generated from the synthetic rendering / reconstruction pipeline; no personal data is annotated. |
+| Is data compliant with data subject requests for data correction or removal, if such a request was made? | Not Applicable — no personal data. |
+| Applicable Privacy Policy | <https://www.nvidia.com/en-us/about-nvidia/privacy-policy/> |
+
+##  Safety
+
+| Field | Response |
+|---|---|
+| Dataset Application Field(s): | Robotics Learning; Humanoid Whole-Body Control; Loco-Manipulation Research; Physics-Based Animation; Sim-to-Real Transfer Research. |
+| Describe the life critical impact (if present). | Not Applicable for direct dataset use. Downstream policies trained on GRAIL motions and deployed on a physical Unitree G1 are operating a hardware actuator and require independent safety validation by the integrating team (torque / velocity limit checks, emergency-stop integration, environment-specific risk assessment). The dataset itself does not encode hardware safety constraints. |
+| Use Case Restrictions: | Abide by the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). GRAIL must not be used to: (1) train policies for deployment on humanoids in public-facing safety-critical roles (e.g. interacting with vulnerable populations) without independent safety validation; (2) produce or distribute deepfake content of real individuals — the dataset contains no real-person data, so any such use would require off-pipeline data that violates the upstream synthesis constraints; (3) violate the licenses of upstream third-party assets (object meshes, character libraries, motion sources). |
+| Dataset restrictions: | The Principle of Least Privilege (PoLP) is applied. Source assets (RoboCasa-derived 3D objects, synthetic character library, motion-source mocap) are tracked with NSpect IDs in the upstream pipelines. The release pipeline strips path-level provenance from the original cluster filesystems before publication. |
+| Security considerations: | The dataset is composed of pickled trajectories (Python `.pkl`), MP4 videos, OpenUSD assets, and a CSV manifest. Pickled files execute arbitrary Python on load; users should verify the integrity of downloaded files (HF hashes / commit signatures) and load them inside a trusted environment. The release ships no executable code, no model weights with auto-execute capability, and makes no network calls when loaded. Report security vulnerabilities to NVIDIA [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/). |
+| Responsible AI practices: | GRAIL is designed to advance public research on humanoid loco-manipulation. Users deploying derived policies on physical humanoids are responsible for hardware-side safety review (joint-limit / torque / E-stop / human-proximity safeguards) prior to deployment. NVIDIA encourages developers to implement validation harnesses, sim-to-real gap analysis, and conservative envelope checks in any pipeline that takes a GRAIL-trained policy to hardware. |