Skip to content

For AI labs · Humanoids and legged robots

Human demonstration data for humanoid robots

Metagaming collects human demonstration data for humanoids and legged robots from vetted adult trainers: balance, dodge-timing and movement demos in games, simulations and VR, plus human judgments on your robot's runs. It fits teams building humanoids and robot dogs, and every project starts as a small pilot.

Request a dataset or pilot Data formats and specs

What's the data problem for humanoids?

A humanoid has to do many things a person does without thinking: keep its balance, step around an obstacle, react in time when something moves, and work through a household task in a sensible order. Each of those needs examples of people doing it well.

Humanoid companies are investing heavily in human data. Figure, for example, said its Series C would fund "data collection of human video and multimodal sensory inputs" (Figure, 2025). Real whole-body data from motion capture or teleoperation is valuable, but slow and costly to collect at volume.

Games, simulations and VR let trainers produce timing, path and sequencing examples quickly. The honest limit: they record a trainer's inputs and the game's state, not joint-level body motion, so they complement motion capture, video and teleop data rather than replace it.

Which Metagaming data fits humanoids?

All four data types open for pilots apply to humanoids and legged robots, and teleoperation is coming soon.

  • Pilots open

    Human demonstrations

    Trainers move, dodge or work through a task in a game, simulator or VR. You get their inputs, timestamps and the task state your brief lists.

  • Pilots open

    Gameplay input data

    Inputs and timing from skilled players in fast movement and VR games, recorded only in the games a mission names.

  • Pilots open

    Evaluation labels

    Trainers judge your robot's runs: did it recover its balance, did it react in time, and which of two runs moved more naturally.

  • Pilots open

    Custom training games

    We build a game around your task, such as stepping through clutter, so trainers produce exactly the demonstrations your model needs.

Coming soon

Teleoperation sessions

Trainers at Expert level and above will control real robots remotely, live, under your instructions. Most teleop work elsewhere is done on site; tell us your robot in the request form to be first in line.

What could a humanoid pilot look like?

These are examples of what a pilot could look like, not past projects. Your brief sets the real scope.

  • Dodge-timing demos

    Trainers react to moving objects in a custom game, stepping aside or stopping in time, with each reaction and its timing recorded.

  • Movement through clutter

    Trainers steer a character through rooms and doorways in a simulation, choosing paths a humanoid or robot dog could follow.

  • Balance and recovery judgments

    Trainers review clips of your robot's runs and label each one, or choose which of two runs recovered better.

Which trainer skills does it use?

Humanoid missions draw on trainers with strong scores in timing, speed, navigation and steady hands, four of the six skills we measure per trainer. Scores come from our skill tests, such as the timing test and the reaction time test, and from quality review on missions.

We recruit skilled gamers, including FPS and VR players, and people with real-world trades whose everyday movement is the skill, such as warehouse, kitchen and home workers. Where a mission needs a certificate, we check it before that mission.

You can filter trainers by level, skill score and industry. Labs see a trainer name, skills and scores, never a real name.

How does a pilot work?

A pilot follows the same five steps as every Metagaming project, at a small size, so you can check fit before a larger order.

  1. 01

    Brief

    You tell us the robot, the task, the data type, the volume and the timeline.

  2. 02

    Custom mission

    We design the mission or training game and agree the QA rules with you.

  3. 03

    Collection

    We match the mission to accepted trainers with the right level, skill scores and industry.

  4. 04

    QA

    Every task is checked against the agreed rules. Work that fails isn't delivered.

  5. 05

    Delivery

    You receive the data in the agreed format, with each session's consent record.

Humanoid robot data: questions from labs

Does this data include full-body motion?

No. Game, sim and VR demonstrations record a trainer's inputs, timing and the task state your brief lists, not joint-level body motion. They work best for timing, paths and task order, alongside motion capture, video or teleop data.

Does it cover legged robots as well as humanoids?

Yes. Movement, timing and recovery data applies to robot dogs and other legged robots as well as humanoids. Tell us your robot in the brief.

Can trainers judge our robot's runs?

Yes. Evaluation labels are judgments on your robot's attempts: what went right, what went wrong, and which of two runs was better. The questions and allowed answers are agreed in the brief.

What format do we get?

The format is agreed in the brief. LeRobot, RLDS, HDF5, MCAP and JSON or CSV are available on request. See data formats and specs.

We're looking for the most skilled people to train the future.

Think you're one of them? Prove it.

Apply free

Free to apply · 18+ · Every application is reviewed