STEP 01
WHAT WE STANDARDIZE
Bring in the work
Collect the real-world human actions, spaces, tools, and failure cases robots need to learn.
→DOMAIN LABS · ROBOTICS · THE REAL WORLD IS THE TRAINING SET
The real world is the training set.
Robotics companies tell us what their robots need to learn. AuraOne finds the right people, captures the right tasks, checks the data, and delivers it for training.
STARTER MODEL
OpenVLA
Open vision-language-action model handoff when the program supports it.
REVIEWER
Robotics safety
Every session reviewed before it becomes training data.
EXPORTS
RLDS · OpenX
HDF5, BVH, and JSON also ship with checksums and signed manifests.
HOW IT WORKS
Three steps. No architecture lecture.
Bring in the work. Standardize the session. Hand off a training-ready dataset.
STEP 02
WHAT WE REVIEW
Standardize the session
Robot run in. Safety review on questionable sessions. The slow, risky, or inconsistent step becomes one clear workflow with review and sign-off attached.
→STEP 03
WHAT WE SIGN
Leave with a clean dataset
Training-ready dataset out. Raw files, task data, reviewer decisions, signed manifests, and checksums ship together.
DATASET FLOW · FOUR STEPS
From skill gap to training-ready data.
Robotics teams define the skill. AuraOne turns it into task briefs, finds the right people and places, checks each session, and packages the accepted data for training.
STEP 01
Find the right people
Every task maps to the people and environments it needs. Homes, kitchens, warehouses, factories, expert skill holders.
→STEP 02
Capture the real task
Phones, cameras, depth, rigs, or teleop when the program requires it. The capture plan stays tied to the robot skill.
→STEP 03
Check the quality
Every session is reviewed before it becomes training data. Accept, rework, or reject — with the reason attached.
→STEP 04
Deliver the dataset
Raw files, task data, reviewer decisions, accepted clip list, signed manifests, and checksums travel together.
SCOPE · WHAT & WHERE
The task tells us what to collect. The world supplies the data.
WHAT WE COLLECT
The kinds of demonstration data robotics teams need before a policy can be trusted in the real world.
Household tasks
Workplace motion
Object handling
Expert demonstrations
Environment walkthroughs
Teleoperation sessions
Robot failure cases
Edge-case examples
WHERE WE COLLECT
The environments where the real tasks happen — not a studio reconstruction, not a synthetic floor.
Homes
Kitchens
Restaurants
Warehouses
Hotels
Retail stores
Factories
Labs
TASK BRIEF BUILDER
Every clip starts with a task brief.
Task briefs tell operators what to record, what environment is needed, what tools or objects matter, how to frame the session, and what causes rework. The brief travels with the clip.
HOUSEHOLD TASK DATA
Fold a bath towel on a kitchen table
Pick up a bath towel, fold it in thirds, then stack it neatly.
ENVIRONMENT
Kitchen or laundry area with a clear table
ACCEPTANCE
Full task visible, no fast cuts, towel edges and hand motions in frame.
HOUSEHOLD TASK DATA
Load dishes into a dishwasher
Open the dishwasher, place plates and cups, adjust one item, close the rack.
ENVIRONMENT
Home kitchen with dishwasher access
ACCEPTANCE
Object placement and drawer motion are visible from start to finish.
WORKPLACE MOTION DATA
Pick oddly shaped grocery items from a shelf
Select irregular items, rotate them, and place them into a tote.
ENVIRONMENT
Shelf, pantry, stockroom, or retail-like setup
ACCEPTANCE
Each grasp includes approach, contact, lift, carry, and placement.
FAILURE CASE DATA
Recover from a dropped object
Handle an object, let it slip safely, pause, recover it, and reset the task.
ENVIRONMENT
Safe household or workplace surface
ACCEPTANCE
Drop, reaction, recovery path, and reset are all visible.
STARTS FROM
A model already suited to the workflow.
STARTER MODEL
OpenVLA
Starts with OpenVLA, LeRobot-style policies, and open robotics models suited to session review and dataset curation.
Operator feedback, review decisions, and downstream training results improve the data program around your robot work.
AuraOne can help stand up the first safety loop. Your robotics team still owns what becomes training data.
READING · ROBOTICS · LIVE
00·00 INTAKESAFETY REVIEWSIGN 04·18
ROBOTICS STUDIO
Four ways to run the same loop.
Review locally, scale to the cloud, deploy inside your tenant, or hand the second pass to managed reviewers. The dataset stays yours.
OPEN · TIER
№ 01
Robotics Studio Open
Free local-first review IDE for teleop and VLA datasets.
STARTS FROM
OPENVLACLD · TIER
№ 02
Robotics Studio Cloud
Hosted multi-reviewer queues, dataset storage, dashboards, and approval chains.
STARTS FROM
OPENVLAENT · TIER
№ 03
Robotics Studio Enterprise
Self-hosted or VPC deployment with SSO, audit-grade evidence, and signed export attestations.
STARTS FROM
OPENVLAPRG · TIER
№ 04
AuraOne Robotics Programs
Managed reviewer pool and dataset request intake for failure annotation, intervention tagging, and training-mix curation.
STARTS FROM
OPENVLACAPTURE NETWORK · FIVE TIERS
Start with what you have. Earn into higher tiers.
Operators record real tasks, get reviewed, and get paid for accepted clips. Tiers move from a phone in a home kitchen all the way to teleop sessions in a robotics cell — gated by program scope and provider setup.
TIER 1 · № 01
Everyday Operators
Home chores, simple object handling, phone capture.
laundry · dishes · pantry sorting · simple object handling
TIER 2 · № 02
Workplace Operators
Kitchens, warehouses, retail, hotels, facilities.
stocking · drawer motion · facility walkthroughs · tote handling
TIER 3 · № 03
Expert Operators
Tools, lab workflows, medical/surgical equipment, industrial tasks.
tool use · lab bench work · industrial process · specialty equipment
TIER 4 · № 04
Teleop Operators
Remote operation and robotics control sessions.
robot control · trajectory capture · reset handling · operator feedback
TIER 5 · № 05
Field Operators
Robot setup, environment walkthroughs, deployment support.
site walkthroughs · robot setup · field notes · environment mapping
TRAINING DELIVERY
Approved sessions become a delivery package.
Delivered datasets include raw files, metadata, review decisions, accepted clip lists, manifests, checksums, and supported training packages. No reconstruction needed.
№ 01
Raw videos
№ 02
Task data
№ 03
Device metadata
№ 04
Environment metadata
№ 05
Operator and session metadata
№ 06
Review status
№ 07
Accept, reject, and rework reasons
№ 08
Accepted clip list
№ 09
Export manifest
№ 10
Checksums
№ 11
Training format package
↳ NATIVE FORMATS
RLDSOpenXHDF5BVHJSONREADING · DELIVERY · LIVE
DELIVERY PROOF · SIX NODES
Not random clips. Robot training data.
Anyone can pay people to record videos. AuraOne helps robotics teams collect the right data — the task, the context, the quality check, the reviewer decision, and the delivery package.
01
Review
Human reviewers check the task, context, quality, and release decision before delivery.
02
Regression memory
Rejected clips can become failure examples that help teams avoid repeating bad behavior.
03
Safety review
Risky sessions can be held for a safety lead before they enter the training set.
04
Export formats
RLDS, OpenX, HDF5, BVH, and JSON through the public export surface.
05
Provider-gated training
Fine-tuning and weights delivery stay gated on provider setup and pilot scope.
06
Delivery package
Raw files, task data, review decisions, signed manifests, and checksums ship together.
HONESTY
Your data stays yours.HONESTY
Workers consent before each session.HONESTY
Rejected clips can become failure examples. Not public datasets.BUILT FOR
Teams turning raw real-world data into robot skill.
From autonomous vehicle labs to warehouse automation teams, the same evaluation surface scales across company types and fleet sizes.
↳ TEAM TYPE
Humanoid robotics teams
Teams collecting human movement, household tasks, workplace motion, and failure examples for physical AI programs.
↳ TEAM TYPE
Autonomy research labs
Research groups that need reviewed raw video, movement data, teleop sessions, and export packages before training changes.
↳ TEAM TYPE
Embodied AI programs
Teams building vision-language-action policies from real people doing real tasks — not lab reconstructions.
↳ TEAM TYPE
Teleoperation research groups
Groups that need structured teleoperation sessions and reviewer decisions for downstream training.
CATEGORY MAP · WHERE WE SIT
What this is and is not.
A short read on the landscape. Anonymized by category, not company — because the point is the work, not the logo.
Fast-capture worker networks
Fast raw clips. AuraOne adds the task design, quality check, and delivery package robotics teams need before training.
Generic gig-labor marketplaces
Massive worker supply, no robotics knowledge. AuraOne is built specifically for robot training data and reviewer-graded sessions.
Open public datasets
Useful starting points. AuraOne helps you build your own proprietary dataset on top of the public baseline.
Internal operations
Spreadsheets, folders, and Slack threads do not scale a robotics data program. The work needs one record, one reviewer queue, one delivery format.
TRUST & SECURITY · WHAT TEAMS ASK
Serious data, simply handled.
QUESTION
Who owns the data?
Your data stays yours. We do not resell customer clips. The accepted clip list, manifests, and weights you tune all belong to your program.
QUESTION
Do workers consent?
Workers consent before each session and see the task they are being asked to capture. The brief is part of the record.
QUESTION
What happens to rejected clips?
Rejected clips are not wasted. They can become failure examples for your robotics team — not public datasets.
QUESTION
How are exports delivered?
Exports are packaged for robotics teams with raw files, task data, review decisions, signed manifests, and checksums attached.
QUESTION
Can you work with our existing teleop stack?
Teleop sessions are available when the customer program, provider setup, and physical environment support them. The capture plan stays scoped to the pilot.
ON THE RECORD · AN EMBODIED AI PROGRAM
“The dataset showed up with the reviewer’s notes still attached to the rejected clips. That’s the part we’d been missing for years.”
WHAT YOU KEEP
Your work. Your data. Your AI.
WORKFLOW
Real sessions
Homes, kitchens, warehouses, factories, tools, and failure cases your robotics team needs.
DATA
Your tenant
In your VPC. Your keys. Your retention policy. Your data stays yours.
WEIGHTS
Yours to keep
Reviewed robotics data, task context, accepted clip lists, manifests, and supported export packages.
RELATED LABS
Same loop. Different wavelength.
ROBOTICS
Bring the workflow you want to own.
We'll map the workflow. Pick the starting model. Standardize the session. Hand you the result.
↳ STARTS FROM
OpenVLA
↳ LEAVES WITH
Reviewed robotics data, task context, accepted clip lists, manifests, and supported export packages.