Praana Labs Talk to us

Platform Praana Xchange

One sentence in. Verified, trainable physical AI data out.

A lab names a gap. Capture is aimed at exactly that, and nothing comes back until physics agrees.

How an order runs

Fig. 01 — The loop, end to end
  1. Intake

    The swarm

    Eight billion phones, watches and headsets are already pointed at the world. Nobody had addressed them as one supply.

    Handheld and wearable device base: IoT Analytics, 2026.

  2. The order

    One sentence in

    A lab writes down what its model still cannot do. That sentence is the order.

  3. The split

    One order becomes many small jobs

    The orchestrator breaks it into tasks a person can do in a few minutes, and sends each one to somebody already carrying the right sensor.

  4. The gate

    Physics decides

    Every return is judged against physics before it counts. A fabricated capture earns nothing.

  5. The suite

    Solved and graded

    Footage is resolved to metric 3D and given a quality tier. The suite is sold on its own, and it will run on a corpus you already hold.

  6. Delivery

    Verified data out

    Training-ready episodes with provenance attached, days after the sentence rather than quarters.

  7. Evals

    Models scored against physics

    The verifier that grades a capture will also grade what a model generates, including models we had no part in training.

  8. Coverage

    What is still thin

    The dimension moderator holds the space of variation the order implies and points the next wave at the regions still missing.

Verification

Every capture is proved before it counts.

Every return crosses two gates the moment it lands, ours and the physics. No human reviews a capture, because none is needed.

Fig. 02 — The verification gauntletIllustrative

What you can order

You order a capability. We work out how to capture it.

Point at one to see the ways it can be captured.

What you need

Hands & objects
Articulated objectsQuests · Teleop · In-situ
Manipulation primitivesReflex · Quests · Teleop
Deformables & granularReal-task · In-situ
Contact-rich forceReal-task · Teleop
Egocentric hand-objectReal-task
Scenes & geometry
Posed indoor sweepsGuided capture
Depth + RGB pairsGuided capture
Object 360° scansGuided capture
Behaviour over time
Failure & recoveryQuests · Real-task · In-situ
Long-horizon multi-stepReal-task · Teleop · In-situ
Locomotion & navigationQuests
Judgment & proof
Human-intuition labelsAmbient judgment
Verification-as-a-serviceRuns on a corpus you already hold

How it gets captured

T0Ambient judgmentNo gear. Seconds.
T1Reflex breaksA bare phone, one motion.
T2Quest sessionsChaptered minutes.
T3Guided captureA coverage bar you fill.
T4Real-task captureGear on, real task.
T5TeleoperationDrive a robot elsewhere.
T6DirectedAn overlay across all tiers.
T7In-situ operationalInside the workplace.
T8Longitudinal & pairedMonths, or human and robot.
Anyone, anywhereGear-gated and site-gated
Your own corpus

Point at what you need. We come back with the route.

Send us the gap
NextContribute captureYou already own the instrument. See what a capture earns and how it settles.
Read on →