GRTraining data for Physical AI & LLMs
Real-world data for machines that learn.
Graviti captures, annotates and licenses production-grade datasets for humanoids, robots and large language models — with consent and provenance built into every record.
Real-world capture
- Egocentric video
- Multi-view video
- Stereo + IMU
- Camera calibration
- Action segments
- Body keypoints
- Custom collection
- LLM data programs
01Flagship dataset
Ego-Exo Workplace Tasks. Synchronized first-person and multi-view video of real work in factories, hotels and offices.
Real workers doing their real jobs — small-part assembly and packaging on factory floors, housekeeping in hotel rooms, tidying and cleaning in offices and reception areas — recorded by a head-mounted camera and three fixed cameras at the same time. Sessions are synchronized and calibrated. Annotators describe each action in free text with timestamps, and a pose model estimates body keypoints; raw footage can also be licensed without annotations.
- Hours of video
- 4,300+
- Synchronized views
- 4
- Frame rate
- 30 fps
- Hours per participant
- ≤ 10
02Real work, recorded in sync
Recorded where the work actually happens.
We don't stage data in labs. Real workers record their own jobs — on factory floors, in hotel rooms and offices — with a head-mounted camera and fixed cameras running in sync, so your models learn from the tasks they will meet in deployment.
03Catalog
Datasets you can license today.
Off-the-shelf datasets for Physical AI and LLMs, each with a public specification and an evaluation sample.
Physical AI
For humanoids, manipulation and world models.
Ego-Exo Workplace Tasks
Synchronized first-person and multi-view video of real work in factories, hotels and offices.
- Hours of video
- 4,300+
- Synchronized views
- 4
Egocentric Household Activities
First-person video of real people doing real chores in real homes.
- Hours of video
- 10,000+
- Contributors
- 2,400+
Bimanual Teleoperation Trajectories
Synchronized multi-view robot demonstrations with language instructions.
- Episodes
- 120,000+
- Task families
- 180
Large language models
For reasoning, alignment and agents.
Expert STEM Problems
Original research-level science problems with expert-written solutions.
- Subjects
- 4
- Level
- Research
Human Preference Pairs
Pairwise and graded comparisons for RLHF, DPO and reward modeling.
- Comparisons
- 400,000+
- Avg. turns
- 4.2
Agentic Coding Environments
Containerized software and tool-use tasks with automated verifiers.
- Task format
- Harbor
- Verification
- Automated
Professional Work Tasks
Expert-graded finance, legal and office tasks with real deliverables.
- Domains
- Finance · Law · Office
- Deliverables
- XLSX · DOCX · PPTX
04How we build data
From the real world to your training run.
One accountable pipeline, from capture hardware to delivered shards. The same pipeline powers our catalog and custom programs.
- 01
Capture
A vetted global network of contributors, operators and domain experts records data on calibrated devices — in homes, workplaces and labs.
Head-mounted cameras · teleop rigs · expert authors
- 02
Annotate
Trained annotators and model-assisted tooling add action segments, body and hand keypoints, language, preferences and rationales.
Hand pose · actions · narration · rubrics
- 03
Verify
Multi-pass human review, automated checks and contamination screens. Every batch ships with measured quality metrics.
Two-pass QA · agreement metrics · audits
- 04
Deliver
Clean rights and provenance, delivered in the formats your training stack already reads — to your bucket, on your schedule.
LeRobot · RLDS · WebDataset · Parquet · JSONL
05Delivery
Ready for your stack on day one.
Standard formats, documented schemas and data cards. Delivered to S3, GCS, Azure or Hugging Face — no conversion scripts required.
- MP4
- YAML
- JSON
- Parquet
- LeRobot on request
ego-exo-workplace/ # indicative; final layout confirmed per delivery
├── manifest.csv
└── sessions/<scene>/<session_id>/
├── ego/
├── exo_cam1/
├── exo_cam2/
├── exo_cam3/
├── calibration.yaml
├── keypoints_2d.json
├── keypoints_3d.json
├── actions.json
└── metadata.json06Rights & compliance
Data you can actually ship a model on.
Clean rights are not an afterthought. Read how we license data on our company page.
Consent by design
Every contributor signs a written, revocable agreement that explicitly covers AI training.
Provenance per record
Each clip, episode or example traces back to its capture session, device and consent record.
Privacy protection
Faces, screens and documents can be blurred before delivery; audio is only included when licensed.
Clear commercial license
Perpetual rights to train and deploy models, with exclusive options for strategic partners.
07FAQ
Questions we hear most.
What is Graviti?
Graviti is a training data company that licenses datasets for Physical AI — including synchronized egocentric and multi-view video of people doing real work in factories, hotels and offices — and for large language models, and builds custom data programs for both.
What datasets does Graviti offer?
Physical AI: Ego-Exo Workplace Tasks (available), Egocentric Household Activities (available) and Bimanual Teleoperation Trajectories (early access). LLM: Expert STEM Problems (early access), Human Preference Pairs (available), Agentic Coding Environments (early access) and Professional Work Tasks (early access). Every catalog dataset has a public specification at graviti.com/datasets and an evaluation sample. Custom collection and LLM data programs are built to order.
Does Graviti provide training data for humanoid robots and robotics?
Yes. Graviti licenses synchronized first-person (egocentric) and multi-view video of real work — factory assembly and packaging, hotel housekeeping, office and reception tasks — with camera calibration, timestamped action segments and model-estimated body keypoints, alongside egocentric household video and robot teleoperation demonstrations. The data is used to pre-train vision-language-action (VLA) models and to train policies for humanoids and robot arms, and Graviti captures new data to order.
Does Graviti provide data for training large language models?
Yes. Graviti licenses original research-level STEM problems with expert-written solutions, containerized coding and tool-use environments with automated verifiers, and rubric-graded finance, legal and office tasks with reference deliverables. All three suit reinforcement learning and evaluation, and Graviti produces custom LLM data for the domains and languages you need.
Can I get a free sample before licensing a dataset?
Yes. Each dataset page has a sample request form. Samples are free and provided for internal evaluation only. Where a ready-made sample exists it is delivered as a download link after a short request with a work email address; otherwise the data team prepares a sample pack for your evaluation.
Get started
Tell us what your model needs to learn.
Share your target tasks, modalities and volume. We reply within one business day with matching datasets, samples and a quote.