Add dataset card and research license

Replace the previous card with one built from the technical report, and add
the ACE-Data-0 Research License Agreement referenced by the gated access form.

- Gated access form (academic research only, manual review) in the card metadata
- viewer: false, since no data files are published yet
- Figures are linked from the public project page rather than stored in this
gated repo, where resolve/ URLs are not readable by unapproved visitors

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

LICENSE

@@ -0,0 +1,173 @@
+ACE-Data-0 RESEARCH LICENSE AGREEMENT
+Version 1.0
+
+Copyright (c) 2026 S-Lab, Nanyang Technological University, and ACE Robotics.
+All rights reserved.
+
+This Agreement governs access to and use of ACE-Data-0. By requesting access to,
+downloading, or using the Dataset, you accept this Agreement and agree to be
+bound by it. If you do not accept it, you may not access or use the Dataset.
+
+
+1. DEFINITIONS
+
+1.1 "Licensors" means S-Lab, Nanyang Technological University and ACE Robotics,
+    jointly.
+
+1.2 "Dataset" means ACE-Data-0, including all recordings (video, audio, motion
+    capture, tactile), object meshes, calibration and synchronization data,
+    annotations, metadata, documentation, and any part or subset thereof, in
+    whatever form distributed by the Licensors.
+
+1.3 "You" means the individual natural person who has been granted access to the
+    Dataset by the Licensors. This Agreement is entered into by you personally,
+    not by your institution.
+
+1.4 "Non-Commercial Academic Research" means research, teaching, and scholarly
+    activity conducted without any commercial purpose and without any direct or
+    indirect commercial advantage, at or under the supervision of an accredited
+    academic or non-profit research institution. Research funded in whole or in
+    part by a commercial entity, research conducted with a view to commercial
+    exploitation, and any use in or in support of a product, service, or
+    internal business process are NOT Non-Commercial Academic Research.
+
+1.5 "Derivative Material" means any model, model weights, feature representation,
+    embedding, annotation, statistic, rendering, reconstruction, synthetic data,
+    or other output that is produced from, trained on, or derived from the
+    Dataset.
+
+1.6 "Participant" means any natural person appearing in, or whose body, hands,
+    voice, motion, or contact signals were recorded in, the Dataset.
+
+
+2. GRANT OF LICENSE
+
+Subject to your continuing compliance with this Agreement, the Licensors grant
+you a worldwide, royalty-free, non-exclusive, non-transferable, non-sublicensable,
+revocable licence to download, store, reproduce, and modify the Dataset, and to
+create Derivative Material, in each case solely for the purpose of Non-Commercial
+Academic Research.
+
+No other rights are granted. All rights not expressly granted are reserved by the
+Licensors.
+
+
+3. RESTRICTIONS
+
+You shall not, and shall not permit or assist any third party to:
+
+3.1 Use the Dataset or any Derivative Material for any commercial purpose,
+    including product or service development, evaluation, demonstration,
+    benchmarking for commercial claims, or any internal business process of a
+    commercial entity.
+
+3.2 Distribute, publish, transmit, sublicense, sell, rent, lend, host, mirror,
+    or otherwise make the Dataset available, in whole or in part, to any person
+    who has not been granted access individually and directly by the Licensors.
+    This includes shared drives, public or private code repositories, torrents,
+    model hubs, and inclusion in any other dataset or collection. Every
+    collaborator, student, or colleague who requires the Dataset must submit
+    their own access request.
+
+3.3 Attempt to identify, contact, locate, track, or infer private attributes of
+    any Participant, or to link any Participant to any other data source.
+
+3.4 Use the Dataset or any Derivative Material to develop, train, evaluate, or
+    deploy systems for biometric identification, face or gait recognition,
+    emotion inference, surveillance, profiling, or any purpose that would
+    infringe the privacy, dignity, or fundamental rights of any person.
+
+3.5 Use the Dataset or any Derivative Material in any manner that violates
+    applicable law, or in any application intended to cause physical or
+    psychological harm.
+
+3.6 Remove, obscure, or alter any copyright, attribution, or licence notice
+    accompanying the Dataset.
+
+3.7 Circumvent, or assist in circumventing, the access controls applied to the
+    Dataset.
+
+
+4. DERIVATIVE MATERIAL AND PUBLICATION
+
+4.1 You may publicly release Derivative Material, including trained model
+    weights, provided that (a) the Derivative Material does not permit
+    reconstruction of any substantial portion of the Dataset and does not
+    reveal the identity of any Participant; (b) it is released for
+    Non-Commercial Academic Research only; and (c) it is accompanied by a
+    notice stating that it was derived from ACE-Data-0 under this Agreement.
+
+4.2 You may reproduce a limited number of images or short clips from the Dataset
+    in academic papers, theses, posters, and conference presentations for the
+    purpose of illustrating your research results, with attribution to
+    ACE-Data-0.
+
+4.3 Any publication, preprint, technical report, or public artifact that uses
+    the Dataset or any Derivative Material must cite ACE-Data-0 as specified in
+    the accompanying dataset card.
+
+
+5. TERM AND TERMINATION
+
+5.1 This Agreement takes effect when access is granted to you and continues
+    until terminated.
+
+5.2 The Licensors may terminate this Agreement and revoke your access at any
+    time, with or without cause, and immediately upon any breach by you.
+
+5.3 On termination you must cease all use of the Dataset and delete all copies
+    of the Dataset in your possession or control, and confirm this in writing on
+    request. Sections 3, 6, 7, 8, and 9 survive termination.
+
+5.4 The Licensors may modify or withdraw all or part of the Dataset at any time,
+    including at the request of a Participant. Where a Participant withdraws
+    consent, you must delete the affected recordings on notice.
+
+
+6. NO WARRANTY
+
+THE DATASET IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
+FOR A PARTICULAR PURPOSE, ACCURACY, AND NON-INFRINGEMENT. THE LICENSORS DO NOT
+WARRANT THAT THE DATASET IS FREE OF ERRORS OR THAT IT IS SUITABLE FOR ANY
+PARTICULAR PURPOSE.
+
+
+7. LIMITATION OF LIABILITY
+
+TO THE MAXIMUM EXTENT PERMITTED BY LAW, IN NO EVENT SHALL THE LICENSORS BE LIABLE
+FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE DATASET OR
+THE USE OF OR OTHER DEALINGS IN THE DATASET.
+
+
+8. INDEMNITY
+
+You shall indemnify and hold harmless the Licensors and their officers,
+employees, students, and agents against any claim, loss, damage, cost, or expense
+arising from your use of the Dataset or from your breach of this Agreement.
+
+
+9. GENERAL
+
+9.1 This Agreement is governed by the laws of Singapore, and the courts of
+    Singapore shall have exclusive jurisdiction over any dispute arising from it.
+
+9.2 This Agreement constitutes the entire agreement between you and the Licensors
+    concerning the Dataset and supersedes any prior understanding.
+
+9.3 If any provision is held unenforceable, the remaining provisions remain in
+    full force.
+
+9.4 Failure by the Licensors to enforce any provision is not a waiver of that
+    provision.
+
+
+10. CONTACT
+
+Requests concerning commercial licensing, industrial collaboration, use outside
+the scope of this Agreement, or participant data withdrawal should be addressed
+to the ACE-Data-0 authors:
+
+    <CONTACT EMAIL: TO BE FILLED IN>
+    Project page: https://ace-data-engine.github.io/ACE-Data-0/

README.md

@@ -0,0 +1,496 @@
+---
+pretty_name: ACE-Data-0
+license: other
+license_name: ace-data-0-research-license
+license_link: LICENSE
+viewer: false
+language:
+  - en
+task_categories:
+  - robotics
+  - keypoint-detection
+  - video-classification
+  - audio-classification
+size_categories:
+  - 10K<n<100K
+tags:
+  - video
+  - audio
+  - 3d
+  - timeseries
+  - robotics
+  - embodied-ai
+  - multimodal
+  - egocentric
+  - multi-view
+  - motion-capture
+  - tactile-sensing
+  - human-object-interaction
+  - human-scene-interaction
+  - long-horizon
+  - smpl-x
+  - mano
+  - imitation-learning
+  - vision-language-action
+extra_gated_heading: Request access to ACE-Data-0
+extra_gated_description: >-
+  ACE-Data-0 contains identifiable recordings of human participants captured
+  inside real homes. Access is granted to named individuals for non-commercial
+  academic research only, and every request is reviewed manually.
+extra_gated_prompt: >-
+  **Before requesting access, please read the following.**
+
+
+  ACE-Data-0 is released **exclusively for non-commercial academic research**
+  under the [ACE-Data-0 Research License
+  Agreement](https://huggingface.co/datasets/acerobotics2025/ACE-Data-0/blob/main/LICENSE).
+  By submitting this form you confirm that you have read that agreement and that
+  you agree to be bound by it.
+
+
+  The dataset contains **video, audio, motion, and tactile recordings of
+  identifiable human participants** who volunteered and gave informed consent
+  for research release. In particular, you agree that you will **not**:
+
+
+  - use the data, or anything derived from it, for any commercial purpose;
+
+  - redistribute, publish, or otherwise share the data with anyone who has not
+  been granted access individually through this form;
+
+  - attempt to identify, contact, locate, or infer private attributes of any
+  participant appearing in the recordings;
+
+  - use the data to develop or evaluate biometric identification, surveillance,
+  or profiling systems.
+
+
+  Access is granted to **you personally and is not transferable**. Every
+  colleague, student, or collaborator who needs the data must submit their own
+  request. Requests submitted with personal email addresses, without a verifiable
+  institutional affiliation, or without a specific research plan will be declined.
+
+
+  Commercial licensing, industrial collaboration, and any use outside the scope
+  above are handled separately. Please contact the authors rather than
+  submitting this form.
+extra_gated_fields:
+  Full name: text
+  Institutional email address: text
+  Institution or organization: text
+  Country: country
+  Position:
+    type: select
+    options:
+      - Undergraduate student
+      - Master's student
+      - PhD student
+      - Postdoctoral researcher
+      - Faculty / Principal investigator
+      - Research scientist / Research engineer
+      - label: Other
+        value: other
+  Homepage, Google Scholar, or lab page: text
+  Name and email of your academic supervisor (students only, otherwise write N/A): text
+  Which parts of ACE-Data-0 do you intend to use?:
+    type: select
+    options:
+      - Egocentric video
+      - Exocentric video
+      - Human body and hand motion
+      - Object meshes and 6-DoF poses
+      - Audio
+      - Tactile
+      - Annotations only
+      - The complete dataset
+  Describe your intended research use in at least two sentences (project, tasks, and expected outputs): text
+  I confirm that I am requesting access for non-commercial academic research only: checkbox
+  I have read and agree to the ACE-Data-0 Research License Agreement: checkbox
+  I will not redistribute the dataset or any part of it, and I will direct colleagues to submit their own request: checkbox
+  I will not attempt to identify participants, nor use the data for biometric identification, surveillance, or profiling: checkbox
+  I agree to cite ACE-Data-0 in any publication or public artifact that uses it: checkbox
+  I understand that access is personal, non-transferable, and may be revoked: checkbox
+extra_gated_button_content: Submit access request
+---
+
+<div align="center">
+
+# ACE-Data-0
+
+## Human-Centric Ambient Capture as Embodied Data Engine
+
+**S-Lab, Nanyang Technological University, Singapore** &nbsp;·&nbsp; **ACE Robotics**
+
+[![Blog](https://img.shields.io/badge/Blog-ACE--Data--0-0054A6?logo=googlechrome&logoColor=white)](https://ace-data-engine.github.io/ACE-Data-0/)
+![Technical Report](https://img.shields.io/badge/Technical_Report-Coming_Soon-6B7280)
+[![License](https://img.shields.io/badge/License-Research_Only-A33B20)](./LICENSE)
+![Release](https://img.shields.io/badge/Data_Files-Coming_Soon-6B7280)
+
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/data-teaser.webp" alt="ACE-Data-0 teaser: table-scale and room-scale ambient capture with synchronized multi-modal streams">
+
+**ACE transforms real home environments into spatially calibrated, temporally synchronized
+recording studios for embodied AI.**
+
+▶ [Watch the demo video](https://ace-data-engine.github.io/ACE-Data-0/assets/videos/teaser-video.mp4)
+&nbsp;·&nbsp;
+[Read the full story on the blog](https://ace-data-engine.github.io/ACE-Data-0/)
+
+</div>
+
+## Dataset Description
+
+### Dataset Summary
+
+Embodied intelligence faces a fundamental data bottleneck. Learning to act in the physical world
+requires more than observing what an action looks like: models must capture how first-person
+perception, whole-body motion, dexterous manipulation, object state, sound, and touch evolve
+**together** as humans pursue goals over time. Existing datasets typically fragment this experience
+across viewpoints, modalities, or spatial scales, leaving the full perception-action loop only
+partially observed.
+
+We introduce the **A**mbient **C**apture **E**ngine (**ACE**), a human-centric data engine that
+transforms real home environments into spatially calibrated, temporally synchronized recording
+studios. ACE operates at two complementary scales:
+
+- **Table-scale capture** resolves fine-grained hand-object manipulation around work surfaces.
+- **Room-scale capture** preserves whole-body motion, locomotion, and interactions distributed
+  across a furnished home.
+
+Across both settings, ACE records egocentric and multi-view exocentric video, full-body and
+articulated hand motion, high-fidelity object geometry and 6-DoF trajectories, multi-channel audio,
+and tactile signals as a unified multisensory stream.
+
+Using ACE, we build **ACE-Data-0**, a large-scale dataset of natural, goal-directed household
+activity spanning atomic manipulation, long-horizon chains of household activities, and human-scene
+interaction.
+
+### Dataset Scale
+
+| Statistic | ACE-Data-0 |
+| --- | ---: |
+| Recorded activity | 150 hours+ |
+| Video frames | 17M+ |
+| Interaction episodes | 75,000+ |
+| Task categories | 200+ |
+| Participants | 50+ |
+| Capture environments | 2 |
+| Views per moment | 8+ exocentric, plus 4 egocentric fisheye |
+| Take length | minutes, not seconds; up to 20-30 min for long-horizon chains |
+| Raw data rate | approx. 1 TB per hour of session |
+
+These values describe the planned ACE-Data-0 release and will be verified in the final release
+manifest.
+
+An *episode* is a contiguous segment of interaction that realizes one meaningful sub-goal, the
+smallest unit that stays semantically self-contained as a training example. Episodes are counted
+within takes rather than recorded in isolation, so the actions preceding and following each segment
+are preserved in the same stream.
+
+### What Makes ACE-Data-0 Different
+
+- **Nothing is fragmented.** Egocentric video, exocentric video, body and hand motion, object pose,
+  audio, and touch are recorded in one pass, on one clock, in one world frame, not assembled from
+  separate sources.
+- **Real homes, not laboratories.** The room-scale site is a fully furnished 200 m² apartment. The
+  clutter, furniture occlusion, and spatial constraints that lab capture removes are exactly what
+  makes home interaction hard, and they are kept.
+- **Long horizons.** Participants receive **goal-level** instructions ("prepare a cup of tea and
+  serve it at the table") rather than step-by-step scripts. Planning, hesitation, and improvisation
+  enter the data by themselves.
+- **Annotations are measured, not estimated.** Human states, object states, and contact are
+  metrically tracked or directly sensed. Pose reprojections, bounding boxes, motion trails, and
+  contact events follow by projection, with no estimation model in the loop. Only the textual
+  descriptions are generated, and those are human-verified.
+
+## Dataset Structure
+
+### Data Modalities
+
+Each take is organized around a shared timeline and a shared world coordinate frame.
+
+| Modality | Contents | Role |
+| --- | --- | --- |
+| Egocentric video | 4 fisheye views, IMU readings, per-frame 6-DoF headset pose from the tracked rig | Participant-centric perception |
+| Exocentric video | 8 synchronized fixed-camera views, each with intrinsics and world-frame pose | Scene, body, and object context |
+| Human motion | 41-joint body skeletons plus converted SMPL-X parameters | Metric kinematic supervision |
+| Hand motion | Articulated hand state and per-view projections | Dexterous manipulation supervision |
+| Object state | Scanned or 2DGS-reconstructed meshes, identity, 6-DoF pose at 60 Hz, motion trails | Dynamic scene-state supervision |
+| Audio | Synchronized multi-source audio from the exocentric cameras and the headset | Contact, appliance, and ambient events |
+| Tactile | Hand-shaped pressure grids with calibrated normalization and baseline correction | Physical interaction supervision |
+| Calibration | Camera intrinsics, extrinsics, headset pose, world registration, cross-stream timeline | Cross-view and cross-modal alignment |
+| Language | Goal- and activity-level descriptions | Semantic supervision |
+
+Because every take ships its calibration and its timeline as data, any tracked 3D point can be
+projected onto any pixel of any view, and any two streams can be paired at any instant. Modalities
+can therefore be combined freely for training, as inputs or as supervision, without rerunning any
+part of the capture pipeline.
+
+### Annotations
+
+| Type | Contents |
+| --- | --- |
+| Object | category label, 2D/3D bounding boxes in all views, per-frame 6-DoF pose, mesh, motion trail |
+| Human | full-body pose per frame, reprojected as pixel-aligned 2D overlays in every view plus metric 3D in the world frame |
+| Hand | articulated finger configurations during dexterous manipulation |
+| Tactile | per-frame pressure reading, aligned to the visual streams and to the object currently in use |
+| Audio and language | sound-event descriptions, plus the take's goal and the sequence of sub-goals reaching it |
+| Calibration and sync | per-camera intrinsics, camera poses in the shared world frame, full cross-stream timeline |
+
+### Activity Families
+
+| Take type | Content | Typical length |
+| --- | --- | --- |
+| **Atomic HOI** | 1-3 household tasks drawn from 15+ activity types (pouring water, drinking, making tea, watering plants, chopping vegetables, cooking, tidying up) | ~3 min |
+| **Chain of HOI** | the full range of short tasks combined into one continuous activity; sub-tasks interleave freely, and every take ends with the scene tidied back into order | ~20-30 min |
+| **Human-Scene Interaction (HSI)** | almost no objects; whole-body motion (walking, exercising) and human-scene contact (sitting, lying, leaning) with tables, chairs, and sofas | ~5 min |
+
+<div align="center">
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/script-example.webp" alt="Examples of ACE-Data-0 task categories and long-horizon capture scripts" width="88%">
+<br>
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/statistics.webp" alt="ACE-Data-0 task families and dataset statistics" width="88%">
+</div>
+
+## Dataset Creation
+
+### Capture System
+
+|  | Table-scale | Room-scale |
+| --- | --- | --- |
+| Space | ~30 m² workspace around a work desk | ~200 m² furnished apartment (kitchen, dining, living room, bedroom) |
+| Target | fine-grained dexterous hand-object manipulation | whole-body activity, locomotion, scene-wide interaction |
+| Exocentric RGB | 8 × GoPro at close range (0.3-0.5 m), 1920×1080 @ 30 FPS | 8 × ZED One on adjustable poles, 1920×1080 @ 30 FPS, at least 4 views on any point |
+| Optical mocap | 16 × OptiTrack PrimeX 22 | 12 × OptiTrack PrimeX 22 |
+| Hand pose | RANSAC triangulation of 2D keypoints from 8 exo views, plus manual refinement | Manus mocap gloves @ 60 Hz |
+| Objects | 25+ interactable instances, 8+ categories | 25+ interactable instances |
+
+**Shared across both sites**
+
+| Device | Role | Spec |
+| --- | --- | --- |
+| ACE-Ego-Head-V02 Lite | Egocentric capture | 4 fisheye cameras (front/rear, L/R), 1088×1280 @ 20 FPS, onboard IMU, 5 tracked markers |
+| OptiTrack PrimeX 22 | Optical motion capture | 2048×1088 @ 60 Hz; 41 body markers per participant, plus objects and the ego rig |
+| ACE-Sense-Glove Lite | Contact pressure | full-palm pressure map, both hands |
+| GoPro / ACE-Ego-Head | Audio | contact events, appliance operation, ambient scene sound |
+
+<div align="center">
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/setup-ace-t.webp" alt="Table-scale configuration" width="45%">
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/setup-ace-r.webp" alt="Room-scale configuration" width="45%">
+</div>
+
+### Capture Protocol
+
+Participants receive household goals rather than fixed sequences of atomic actions. They choose
+routes, object instances, grasp strategies, subtask order, pauses, and recovery behavior. This
+preserves natural variation while keeping the activity goal and the measured physical state
+available as supervision. Each participant completes all designed tasks across a 2-day session.
+
+Every session follows the same five-step protocol at both sites:
+
+1. **Scene preparation.** Objects are placed at randomized yet plausible initial positions.
+2. **Participant setup.** The participant puts on the mocap suit, the headset, and the gloves, then
+   performs a short T-pose routine that registers their skeleton with the tracking system.
+3. **Task briefing.** The participant receives a goal-level instruction verbally; how to achieve it
+   is left entirely to them.
+4. **Recording.** The take opens with the clock glance described below, after which all sensors
+   record continuously while an operator monitors stream health on a live dashboard.
+5. **Post-checks.** Synchronization and tracking quality are verified after each take, and failed
+   takes are flagged for re-capture.
+
+<div align="center">
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/data-procedure.webp" alt="ACE capture, temporal synchronization, and annotation workflow" width="88%">
+</div>
+
+### Temporal Synchronization
+
+All devices are registered to the OptiTrack 60 Hz clock, which internally delivers strictly
+simultaneous frames. Because recorded camera timestamps lag true exposure by approximately 0.29 s
+and drift slowly, we instead let each camera **photograph a clock**: the mocap host displays its own
+time as a nanosecond-resolution QR code, and readings sampled across a take are fitted with a
+constant offset plus a slow drift.
+
+- Residual alignment error: **millisecond level** (within a single mocap frame).
+- Egocentric fisheye views, mutual misalignment: **< 2 ms**.
+- Independent check, per-camera offsets re-estimated during calibration: **< 4 ms**.
+- Tactile gloves are aligned to the headset by matching correlated IMU motion templates.
+
+Every take ships a table mapping each camera frame, egocentric and exocentric alike, to its 60 Hz
+mocap frame.
+
+### Spatial Calibration
+
+The exocentric cameras never move but barely share a view; the egocentric cameras move every frame.
+Both are solved through the motion-capture system rather than through co-visibility.
+
+- **Exocentric.** An ArUco board with a retroreflective marker at each corner is visible to the RGB
+  cameras and the infrared cameras at once, bridging cameras that share no field of view. Median
+  reprojection error on held-out frames: **< 3 px**.
+- **Egocentric.** The 5 markers on the headset chassis form a rigid body tracked at 60 Hz, so the
+  only unknown is the fixed camera-to-body transform, solved as hand-eye calibration and refined by
+  joint bundle adjustment. Median reprojection error: **approx. 2 px**. Camera poses are therefore
+  *measured* rather than estimated, and do not drift.
+
+Calibration is re-verified twice a day. This alignment lets one modality supervise another without
+uncertain temporal or spatial correspondence.
+
+### Annotation Process
+
+Most annotations are derived from the synchronized and calibrated capture system rather than
+estimated independently from video. Because the human and object annotations come from measured 3D
+states rather than image-based detectors, they remain correct where detectors typically fail: under
+furniture occlusion, extreme viewpoints, and motion blur.
+
+Textual descriptions are the one generated type. Gemini-3.1-pro-preview watches the ego-view video
+and describes each time span in natural language, segment by segment; human annotators then check
+the auto-generated labels and correct the descriptions by hand.
+
+<div align="center">
+<b>Human pose annotation</b><br>
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/annotation-smplx-sg.webp" alt="Human pose annotation across synchronized views and 3D motion" width="88%">
+<br><br>
+<b>Object state annotation</b><br>
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/annotation-obj-sg.webp" alt="Object state annotation across synchronized views and 3D motion" width="88%">
+<br><br>
+<b>Articulated hand annotation</b><br>
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/annotation-hand.webp" alt="Articulated hand annotation" width="88%">
+<br><br>
+<b>Tactile annotation</b><br>
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/annotation-tactile.webp" alt="Tactile annotation" width="88%">
+<br><br>
+<b>Language annotation</b><br>
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/annotation-caption.webp" alt="Language annotation" width="88%">
+</div>
+
+### Source Data
+
+ACE-Data-0 consists of original recordings collected with the ACE capture system. It is not
+assembled from scraped internet video or third-party household datasets.
+
+## Benchmark
+
+<div align="center">
+<img src="https://ace-data-engine.github.io/ACE-Data-0/assets/images/benchmark-steps.webp" alt="Three-level benchmark: signals, scene components, interactions" width="88%">
+</div>
+
+We hold out **10 hours** of capture as a test set and evaluate **30+ published methods** with their
+officially released checkpoints, across three levels that mirror the capabilities an embodied agent
+must chain together: sensing contact, estimating scene state, mastering hand-object coordination.
+
+**1. Low-level signals: tactile from vision.** Predict full-hand grasp pressure from egocentric
+video, evaluated against synchronized glove measurements (temporal accuracy, contact IoU, volumetric
+IoU, center-of-pressure error). *Finding:* detecting **when** contact occurs is far easier than
+recovering **where** pressure is distributed. The best method's absolute C-IoU and V-IoU remain
+modest, and a substantial generalization gap persists.
+
+**2. Scene components: human motion estimation.** 22 methods across per-frame, temporal,
+scene-aware, single-view exo, multi-view exo, and egocentric settings, all against the same metric
+ground truth. *Findings:* (i) strong Procrustes-aligned pose accuracy does **not** imply an accurate
+world-frame trajectory, and the two rankings differ; (ii) scene context mainly helps locate the body
+in the room rather than improve joint configuration; (iii) egocentric methods lag because much of
+the body is out of frame.
+
+**3. Embodied interaction: hand motion from ego and exo views.** The same interactions are recorded
+simultaneously from both viewpoints, so the comparison changes only the viewpoint. *Findings:*
+exocentric methods reach approximately 9.1 mm PA-MPJPE and a 63 mm world-frame trajectory error,
+while egocentric world-space methods remain near 100 mm. This points at **egomotion estimation**,
+not finger articulation, as the dominant error source in egocentric hand reconstruction.
+
+Full tables, metric definitions, and per-method results are in the technical report.
+
+## Considerations for Using the Data
+
+### Intended Uses
+
+ACE-Data-0 is intended for non-commercial academic research on:
+
+- Embodied perception and interaction understanding
+- Human and hand motion recovery
+- Human-object and human-scene interaction
+- Multi-view and egocentric video understanding
+- Cross-modal learning across vision, motion, audio, and touch
+- Object-state estimation and dynamic 3D scene understanding
+- Imitation learning, world models, and vision-language-action systems
+
+### Out-of-Scope and Prohibited Uses
+
+The dataset is not intended for:
+
+- Identifying, re-identifying, or profiling participants
+- Biometric recognition or surveillance
+- Inferring sensitive personal attributes
+- Training systems for harmful, deceptive, or privacy-invasive applications
+- Any commercial purpose
+- Representing all homes, cultures, bodies, abilities, or household practices without further
+  validation
+
+The [ACE-Data-0 Research License Agreement](./LICENSE) is binding and defines the full set of
+restrictions.
+
+### Personal and Sensitive Information
+
+All participants volunteered and signed a written informed consent form covering data collection and
+research release, including the appearance of their faces. The dataset contains identifiable
+individuals. Any attempt to identify, contact, locate, or infer private attributes of a participant,
+and any use of the data for biometric identification, surveillance, or profiling, is prohibited by
+the license and will result in access being revoked.
+
+If you are a participant and wish to have your recordings withdrawn, contact the authors and the
+affected takes will be removed from subsequent releases.
+
+### Known Limitations
+
+- **Site coverage.** Two sites only, hence limited variation in layouts, furnishings, and lighting.
+- **Instrumented ground truth.** Tracked objects must be scanned and marked in advance. State
+  changes of articulated mechanisms, fluids, and deformable materials are not annotated.
+- **Visible instrumentation.** The mocap suit, gloves, headset, and markers appear in the recordings
+  and may introduce dataset-specific visual cues.
+
+## Access and Download
+
+This repository is **gated**. Access is granted to named individuals for **non-commercial academic
+research only**, and each request is reviewed manually, so expect a delay of several days.
+
+1. Sign in to Hugging Face and submit the access form on this page.
+2. Use your **institutional** email and give a concrete description of the intended research.
+   Requests without a verifiable affiliation or a specific research plan will be declined.
+3. Access is **personal and non-transferable**. Collaborators and students must each submit their
+   own request; redistributing the data terminates your license.
+
+For commercial licensing or industrial collaboration, contact the authors directly instead of
+submitting the form.
+
+> **Status.** Data files are not published yet. This card describes the dataset and its capture
+> system; repository organization, storage requirements, checksums, and loading examples will be
+> added at release time. Because the dataset comprises large synchronized media and sensor streams,
+> the release will use sharded archives rather than direct browser downloads. Approved users retain
+> access across the release.
+
+## License
+
+Released under the [ACE-Data-0 Research License Agreement](./LICENSE): non-commercial academic
+research use only, no redistribution, no re-identification. Please read it in full before requesting
+access.
+
+## Citation
+
+The technical report is in preparation. In the meantime, please cite:
+
+```bibtex
+@misc{cao2026acedata0,
+  title         = {ACE-Data-0: Human-Centric Ambient Capture as Embodied Data Engine},
+  author        = {Cao, Yukang and Xie, Haozhe and Wen, Beichen and Yao, Runmao and
+                   Liu, Yinghao and Huang, Yue and Liao, Zhichao and Wang, Yunxiang and
+                   Liu, Haiheng and Tian, Xingshun and Su, Dawei and Zhuo, Long and
+                   Tao, Dacheng and Wang, Xiaogang and Pan, Liang and Liu, Ziwei},
+  year          = {2026},
+  howpublished  = {\url{https://ace-data-engine.github.io/ACE-Data-0/}}
+}
+```
+
+## Contact
+
+For project updates, visit the [ACE-Data-0 blog](https://ace-data-engine.github.io/ACE-Data-0/),
+which hosts the full story, the audio-enabled demo, interactive multi-camera examples, and a Chinese
+version.
+
+Questions about access, licensing, or annotations should be directed to the contact address in the
+[LICENSE](./LICENSE).