Trades Data AI
For robotics & embodied-AI teams

Diverse, license-clean human-demonstration data for contact-rich manipulation.

Egocentric video + IMU + hand-tracking of real skilled manual work, delivered as a LeRobotDataset. Per-field confidence and a MEASURED/INFERRED truth state on every label, cross-stream agreement as the anti-fabrication signal, and a datasheet per clip.

Explore a sample episode →Browse the sample catalog →
How is this much detail real?

We resolve what we can first. Then we ask only what we can't.

The review session is not a survey — it is a reasoning-exhaustion funnel. The system infers, audits, and cross-checks everything it can before it ever asks the worker a question. Only genuinely worker-exclusive information escalates. This is the architecture; the published buyer-facing session populates from the live run once it lands.

System resolved from sensors + code
action: "seat fastener"evidence_span_ms: [161240, 162480]truth_state: INFERRED · confidence: 0.83cross_stream: video↔wrist_imu AGREE
▼ escalated only after exhausting sensor evidence
Residual gap — worker-exclusive
We knowYou picked up the part from the left side of the bench (IMU + video agree, 0.91 confidence).
Surgical gapWas that placement part of your standard setup, or did something earlier change the order?
WHY: UNKNOWN_UNTIL_WORKER_ANSWER
▼ unanswered claims stay UNKNOWN — never shipped as fact
Outcome
truth_state: WORKER_CONFIRMED — if answeredtruth_state: UNKNOWN — if not yet answeredNeither state inflates the other

Architecture diagram — not a count of confirmed labels. Published buyer-facing sessions populate from real worker answers; no confirmed labels ship until a real answer lands.

Scrub a real episode in your browser

The label, the evidence, and the review state — synced to the footage.

Our viewer plays the raw episode alongside the structured layers a model team actually audits: an evidence timeline synced to the video, an artifact-status panel that marks every surface missing / degraded / inferred / confirmed, and a signal-quality panel showing cross-stream agreement as visible anti-fabrication evidence.

HO-Cap public-dataset demo · live

Our pipeline processed the license-clean HO-Cap hand-object dataset (CC BY 4.0, attribution rendered in-viewer). Scrub the episode, inspect each layer, and verify the per-field truth state.

Open the sample episode →
Deployment-readiness data

One action span. Two independent expert reads. One annotated record.

The same audited understanding of each action span carries two expert-grade dimensions: the worker's own account of intent and physical context, and an independent safety / code-compliance read against the codes a robot must meet to operate on a real regulated jobsite. These lanes never cross-contaminate — the audited-linkage panel shows which evidence backs which claim.

Action span — both reads share this anchor
actionseat fastener with impact driver
evidence_span_ms[161240, 162480]
confidence0.83
cross_streamvideo↔wrist_imu AGREE
Worker expert read
Physical context + intent
why_claimWorker-stated intent and situational context for this action
physical_noteGrip variant, body position, load condition — worker-exclusive knowledge
truth_state: UNKNOWN_UNTIL_WORKER_ANSWER

Pending real Clerk-bound worker answer. No confirmed WHY label ships until it lands.

Safety / code-compliance read
Deployment-readiness for regulated jobsites
behavior_quality_labelObserved interaction against the codes a robot must meet on a real regulated jobsite — OSHA, NFPA, applicable trade codes
severity / code_refSeverity tier + specific code citation for the observation
truth_state: INFERRED → pending credentialed expert confirmation

Credentialed safety role is currently the founder. PROFESSIONAL_* truth state reaches confirmed once a credentialed non-founder expert answers.

Architecture diagram. The diagram shows the mechanism and the honest pending state — not a population of dual-confirmed labels. Labels claim the architecture and what will publish; actual confirmed counts render in the live sample once real answers land.

Signal quality

Every field carries its truth state. Agreement is the confidence signal.

We don't flatten uncertainty into a single score. Video↔wrist-IMU agreement is computed per-moment and published as the confidence signal — it dips honestly during occlusion, which is precisely what monocular-only methods can't replicate. Broader multi-modal fusion (phone IMU, audio, scene-OCR) is the build direction; the wrist-IMU agreement layer is what ships today.

Per-field truth states
  • MEASUREDRead directly from a sensor or pixel — no inference.
  • INFERREDDerived from evidence; carries its confidence score.
  • WORKER_CONFIRMEDWorker reviewed + confirmed via the session.
  • DEGRADEDPresent but low quality; signal below threshold.
  • UNKNOWNNot determinable from available evidence.
  • REJECTEDFailed a named gate — never shipped as fact.
Video↔wrist-IMU agreement — the published confidence signal
contact span · fastener seated
0.91 — video↔wrist AGREE
occlusion span · hand obscured
0.61 — video DEGRADED · wrist MEASURED
full occlusion · IMU-only recovery
0.44 — video MISSING · wrist MEASURED (dominant wrist)

Illustrative signal shape — not from a specific clip. Real agreement vectors render in the live sample. During full occlusion, the dominant-wrist IMU continues to produce MEASURED kinematics where monocular pose estimation produces nothing.

Drop-in format

LeRobot v3 / RLDS in three lines. Croissant datasheet per clip.

Every clip ships as a LeRobotDataset (v3 Parquet layout, RLDS-compatible) with per-field confidence and truth-state as sibling columns — so your training loop can filter by epistemic quality without post-processing. A scored Croissant 1.1 datasheet ships with each clip, following the Gebru datasheet format with modality-presence matrix and named gates.

LeRobot v3 / RLDS load snippet
format example — live LeRobot layer renders in the sample once the GPU run lands
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

ds = LeRobotDataset("trades-data-ai/hocap-demo-01")

# Every observation column has a sibling truth-state column:
#   obs.wrist_imu.gyro_rad_s      → MEASURED (direct sensor read)
#   obs.wrist_imu.gyro_rad_s.ts   → truth state enum
#   obs.video.contact_active      → INFERRED (cross-stream agree)
#   obs.video.contact_active.ts   → truth state enum
#   obs.why_claim                 → UNKNOWN (pending worker review)
#
# Filter to only worker-confirmed episodes:
confirmed = ds.filter(
    lambda ep: ep["obs.action_truth_state"] == "WORKER_CONFIRMED"
)

# RLDS-compatible — works directly with Open-X loaders.
Croissant 1.1 datasheet — scored per clip
Modality presence matrix
Egocentric videoPRESENTHelmet-mount, full shift
Wrist IMU (6-DoF)PRESENTMbient, captureGates PASS ≥99.5%
Hand landmarks (21-kp)PRESENTMediaPipe Apache-2.0, buyer-facing
AudioSUPPRESSEDCA two-party consent
Phone IMU / baro / GPSPRESENTPer-consent
3D hand meshINFERREDLicense-clean path (MHR / MoGe); gated
Named gates (clip must clear all to publish)
captureGates PASSWrist IMU arrival-time continuity ≥ 99.5%
cross_stream_alignVideo↔IMU offset measured, confidence > threshold
consent_verifiedSigned agreement on file, PII-cleared
worker_review_completeReview session answered or explicitly deferred
mask_iou_regressionSegmentation mask IoU ≤ 1% worse than accepted baseline

Per-clip datasheet format — live scores render once the GPU run lands.

What the field has found

Your policy doesn't need more demos. It needs more environments.

The published evidence points the same way: diversity of environments and objects generalizes better than sheer demonstration count, teleoperation is throughput-capped, and pooling data across embodiments transfers. We present this as cited industry context — not our own results.

~100
π0.5 approached the performance of a directly-trained baseline after about 100 diverse training environments — environment diversity, not demo count, drove open-world generalization to new homes.
Physical Intelligence, π0.5 (2025)
~50%
RT-1-X lifted mean success ~50% over robot-specific training in the low-data regime by pooling data across 22 robot embodiments (Open X-Embodiment).
Open X-Embodiment / RT-X (2023)
350 hrs / 12 mo
DROID's in-the-wild corpus took 12 months across 13 institutions and 564 scenes to reach ~350 hours — teleop is expensive and throughput-capped.
DROID (RSS 2024)

External research, cited for context. Figures are the authors' reported numbers, not Trades Data AI measurements.

Curation as a product

Publish rate below 100%. Flagged episodes are data, not waste.

Every clip that fails a named gate produces a structured rejection — never a padded result. Flagged episodes carry the gate they failed and the reason, which is itself signal for your eval pipeline (failure-recovery, edge-case distribution). The quality heatmap and per-gate pass/fail render in the live sample; below is the gate taxonomy.

Sensor continuity
Wrist IMU captureGates checked per-packet. Gaps > threshold → QUARANTINED, episode flagged.
structured rejection · not silent discard
Cross-stream alignment
Video↔IMU offset measured with confidence bound. Below threshold → DEGRADED surfaces, not MEASURED.
confidence-graded output · never inflated
Mask IoU regression
Segmentation mask IoU vs accepted baseline. >1% regression blocks publish — forces repair or rejection.
named gate · repair history logged
WHY audit
Every WHY claim audited for evidence traceability. Unevidenced claim → UNKNOWN, not published as INFERRED.
adversarial audit · code citations required
Worker review
Review session must complete or formally defer before a clip reaches buyer-grade. Skipped = not confirmed.
WORKER_CONFIRMED or UNKNOWN · no in-between
Holdout guard
Clips flagged holdout in DB are structurally excluded from every bundle/export — enforced at write path.
DB-enforced · not a policy note
Outcome metrics in your unit — pending

Measured accuracy vs. ground-truth belongs to the first non-founder (MMC pilot) worker run, gated on MANO counsel clearance for the hand-pose layer. Our single founder-recorded methodology clip is n=1 — an accuracy headline from it would be meaningless as a generalization claim. We will publish it honestly when it is real: first non-founder worker, real eval dataset, MMRV / RoboArena rank-correlation cited next to our own figure. No fabricated number ships first.

PENDING: first MMC pilot run + counsel clearance
The unclaimed moat

License-clean, consented, provenance-tracked, human-verified.

Most public ego/manipulation corpora are non-commercial, and acquisition method is now legally decisive. Our differentiator isn't volume — it's data you can actually deploy and defend.

License-clean & consented

Recorded by consenting workers under signed agreements — not scraped. The acquisition method is documented per clip, so you can deploy commercially without a provenance time bomb.

Provenance you can verify

Every published claim links back to the decision and code that produced it, with an in-browser hash recompute. Inspect the chain; don't take our word for it.

Datasheet per clip (Croissant)

A Gebru-style datasheet and Croissant 1.1 metadata ship with each clip, with LeRobot / RLDS load snippets — load it in three lines.

Human-verified truth state

Workers review and confirm labels; cross-stream (video↔IMU) agreement is the published confidence signal. MEASURED / INFERRED / WORKER_CONFIRMED on every field — never confidence theater.

One spine, two scope-tagged products

The same audited understanding emits the robotics dataset and a code-compliance safety read, with no cross-lane leak — the audited-linkage panel shows the lanes per decision.

No fabricated or synthetic-as-real data

Missing renders as missing; a failed gate is a structured rejection, never a padded result. We publish what we don't know.

Verify it yourself in the sample →
Explicit scope

What we sell. What we don't.

Expert buyers trust a vendor that discloses its limits. Here is what we don't sell — which, to this audience, is part of the credibility.

We sell
Contact-rich action-knowledge trajectories (ECoT-compatible)
6-DoF dominant-wrist IMU kinematics (MEASURED)
21-keypoint hand-landmark trajectories fused with wrist IMU (MediaPipe, Apache-2.0)
Per-field truth state + evidence chain per clip
Croissant 1.1 / Gebru-style datasheet per clip
LeRobot v3 / RLDS delivery serialization
Egocentric video of real contact-rich manipulation on real jobsites
Per-clip audited provenance + consent documentation
Code-compliance / deployment-readiness annotations (OSHA / trade codes)
Failure-recovery episodes with named rejection reason
We don't sell
World-model / 4D reconstruction as a product
Scene mesh or dense reconstruction as a product
Robot retargeting
Physics-simulation outputs
Synthetic or generated data shipped as real
MANO-based hand pose in the buyer-facing path (counsel-gated)
Accuracy claims without a real non-founder worker eval
Labels confirmed by models only (no human-in-the-loop bypass)
Get started

Evaluate a sample episode, then the datasheet.

We'll share a sample episode (RLDS / LeRobot-ready) and the per-clip datasheet so your team can judge usefulness before any commitment.

Request access →Browse the catalog →