Skip to content

For AI labs · Autonomous driving

Autonomous driving training data from skilled humans

Metagaming collects autonomous driving training data from vetted adult trainers: driving demonstrations in games and simulators, and human judgments on your planner's edge cases. It fits teams building self-driving cars and delivery bots, and every project starts as a small pilot.

Request a dataset or pilot Data formats and specs

What's the data problem in autonomous driving?

The hard part of driving data is the rare case. Most logged driving is routine, while the moments that decide whether a system is safe, such as a tight merge, a lane closure or unclear right-of-way, are uncommon and risky to stage on a real road.

Teams also need more than trajectories. They need people to say whether a maneuver was acceptable, which of two plans was better, and why. That judgment is hard to script, and it's where skilled human drivers add the most.

Game and simulator demonstrations let trainers drive those rare scenarios on demand, repeat them, and vary them, without anyone on a real road. The trade-off is honest: the vehicle physics are the sim's, not your car's, so this data works best for decisions, timing and judgment, alongside your real-world logs.

Which Metagaming data fits driving?

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

  • Pilots open

    Human demonstrations

    Trainers drive your scenario in a driving game or simulator. You get steering, throttle and brake inputs, timestamps and the sim state your brief lists.

  • Pilots open

    Gameplay input data

    Inputs and timing from skilled players driving in racing and driving games, recorded only in the games a mission names.

  • Pilots open

    Evaluation labels

    Trainers judge your system's runs: what went right, what went wrong, and which of two runs was safer or smoother.

  • Pilots open

    Custom training games

    We build a game around your scenario, such as curbside stops for a delivery bot, so trainers produce exactly the demonstrations you need.

Coming soon

Teleoperation sessions

Trainers at Expert level and above will drive 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 driving pilot look like?

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

  • Merge and lane-change demos

    Trainers drive a set of dense merges and lane closures in a simulator, under your acceptance rules, with each attempt recorded as one episode.

  • Edge-case judgments

    Trainers watch clips of your planner's runs and label each one: acceptable or not, what went wrong, and which of two runs was better.

  • Delivery bot maneuvers

    A custom training game built around low-speed tasks, such as pulling over, parking or passing pedestrians, with your state fields recorded.

Which trainer skills does it use?

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

We recruit skilled gamers, including sim racers, and professional drivers: truck, bus, delivery and rideshare drivers. Where a mission needs licensed drivers, we check the license before that mission. Filtering by license is coming soon, with the first missions that need it.

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 vehicle, the scenarios, 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.

Autonomous driving data: questions from labs

Is this data from real roads?

No. Demonstrations are driven in games and simulators, and evaluation labels are judgments on your system's runs. That makes rare scenarios easy to repeat safely, but the physics are the sim's, so the data works best alongside your real-world logs.

Can the trainers be licensed drivers?

We recruit professional drivers as well as skilled sim racers. Where a mission needs licensed drivers, we check the license before that mission. Filtering by license is coming soon, with the first missions that need it.

Can trainers label our own logs or clips?

Yes. Evaluation labels are judgments on your system'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