STEP 01
WHAT WE COLLECT
Tell us what to capture
Collect the real-world human actions, spaces, tools, and failure cases robots need to learn.
→APP DATA OS · 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.
TWO WAYS TO WORK WITH US
This is the capture side. Real operators recording the real tasks your robot has to learn — on Aura Capture, reviewed and signed. We make the data. You train on it.
HOW IT WORKS
You name it. We go get it.
Tell us what your robot needs to learn. We capture it. You train on it.
STEP 02
HOW WE COLLECT
We capture it
Operators record the real-world tasks on Aura Capture — the spaces, tools, and failure cases your robot has to learn. Every session is reviewed and safety-checked before it counts.
→STEP 03
WHAT WE SIGN
You train on it
Training-ready dataset out. Raw files, task data, reviewer decisions, signed manifests, and checksums ship together.
THE CAPTURE APP
Aura Capture. The app that does the collecting.
Vetted operators record the real-world tasks your robot needs — on their phone, with depth, rigs, or teleop when the program calls for it. They get reviewed, and they get paid for accepted clips. The capture plan stays tied to the skill you asked for.
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.
CAPTURE 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
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.”
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.