The refinery for physical action data

Raw human motion in. Training-ready action data out.

ActiScale is a two-sided marketplace that turns egocentric video from a global collector community into rights-cleared, VLA-ready human demonstration data for robotics labs, refined in six pipeline steps and delivered as LeRobot, RLDS or MP4 + JSON from $8/hr.

A global collector community supplies the raw. An automated pipeline refines it — filtered, segmented, pose- and action-labeled, QC'd, rights-cleared. You get hours your model can train on from day one, not footage you have to clean.

5,000 hrs first release Rights-cleared & consented LeRobot / RLDS-ready

ActiScale at a glance

  • ✓Product: egocentric human demonstration data, refined, QC'd and rights-cleared; first packaged release of 5,000 hours.
  • ✓Price: $8–90 per hour across three tiers, versus $50K–$200K for a single production teleoperation dataset.
  • ✓Formats: LeRobot, RLDS, or raw MP4 with a JSON metadata sidecar.
  • ✓Collectors: record everyday tasks with a phone and harness and are paid per accepted hour.

Why now

Why is demonstration data scarce?

Every serious embodied-AI team needs large, diverse, real-world demonstration data — and there is almost none they can simply buy and license. That gap is why ActiScale exists.

~20K hrs
The entire public robot-teleop supply — demand from frontier labs dwarfs it.
$50K–$200K
Cost of a single production teleoperation dataset today.
5,000 hrs
Our first packaged release — larger than any open teleop set.

Sources: HumanScale / HumanNet (2026); robot-data cost analyses (2026). Figures indicative.

See the data

What does one hour of demonstration data look like?

Real egocentric captures, refined into per-hand mesh, narration and action labels — the structure a model can train on.

Raw egocentric capture

Wide-FOV head-mounted video exactly as recorded — the raw input, before any refinement.

Hand mesh & narration

Per-hand mesh (blue = right, orange = left) tracked on the video, with natural-language narration and action labels.

Tool use & action labels

The same refinement on tool-use tasks across a whole room — hand mesh, narration and per-clip action labels.

Open the live demo →

53 real clips across two capture rigs and 12 task families. Scrub the timeline and watch every annotation layer line up — and see which layer each tier buys you.

The refinement pipeline

How does raw footage become training-ready data?

Almost none can turn what they film into something a model can train on. That refinement is the product — and it is where the cost per delivered hour falls as we scale.

Ingest

Video + sensor streams land with provenance and consent attached.

Filter

Blur, drift, wrong-task and duplicate footage is rejected automatically.

Segment

Continuous footage is cut into discrete, task-aligned clips.

Pose & contact

Hand and body keypoints, object contact events, extracted per frame.

Narrate & label

Natural-language narration and action labels applied to each clip.

QC & package

Tiered peer review, then packaged to LeRobot / RLDS and shipped.

See the refinement pipeline →

How it works

How does the ActiScale marketplace work?

Data → raw in, refined out Settlement ← paid back automatically
Supply

Collectors

Phone + harness + app. Paid per accepted hour.

QC reviewers

Tiered review marketplace. Paid per validated task.

ActiScale

Automated pipeline + settlement

Ingest, refine and QC at scale — then pay the community, automatically.

Demand

AI & robotics labs

Frontier model teams and humanoid-robotics companies.

They pay

Per dataset, per custom order, or by subscription.

Marketplace, not services

We refine and sell hours — not resell labor by the hour.

Automation is the leverage

Agentic AI does more of the filtering, cutting, labeling and QC over time.

Quality-gated

Only work that passes QC is paid — and only work that passes is shipped.

Quality engine

Quality that scales without a management layer.

An automated gate plus a tiered peer-review marketplace. Only work that passes is paid — and only work that passes is shipped.

How we hold the bar →

Rights & compliance

Data you can actually train on.

Consented at source, provenance-tracked, commercial training rights included.

Rights & compliance →

For collectors

How do collectors get paid?

No robot. No studio. A phone, a harness, and the app. You record ordinary human tasks; we pay you for every hour that passes quality review.

Sign up & get the kit

Download the app, mount your phone on the harness, complete a 10-minute calibration task.

Record accepted tasks

Pick tasks from the board — cooking, assembly, cleaning, tool use. Follow the spec, upload.

Get paid per accepted hour

Work that clears QC is paid automatically. Higher accuracy unlocks higher-value tasks.

Earnings come only from work performed — hours recorded, tasks reviewed, audits completed. Never from recruiting.

Join the waitlist

Need demonstration data?

Book a 30-minute data review. We'll scope your task taxonomy and send a free sample slice.

Book a data review Open the live demo

Want to collect?

Join the waitlist. Record everyday tasks, clear QC, get paid per hour.

Join the waitlist