PAC Exchange
Browse datasets

Robot data from the factory floors of Japan.

PAC Exchange is where robotics teams find teleoperation data recorded on real production lines, and where Japanese manufacturers turn that data into a new line of business.

Browse datasets List your data

We focus on the data that is hardest to collect.

Video of people at work is cheap and plentiful. It helps models understand the world, but it cannot show a robot what its own joints and gripper should do.

Data recorded while a skilled operator drives a real robot through a real task can. It is slow and expensive to collect, which is exactly why it is scarce. That is the data we bring to market.

Quality →
PAC focus
1
2
3
4
5
6
7

Scalability (ease of collection) →

  1. Third-person video
  2. Egocentric video
  3. Wearable IMU
  4. Handheld grippers (UMI-style)
  5. Mocap gloves + VR
  6. Leader–follower teleop
  7. Robot teleop on real production lines
Positions are qualitative. Quality here means how directly the data transfers to robot policy training.

Easy to collect

Third-person and egocentric video, wearable sensors. Useful for pre-training, but the robot's actions have to be inferred and transferred to a different body.

Worth collecting

Robots driven by skilled operators with leader–follower arms, gloves or VR. Every frame records the robot's own joint states and actions, on the task you care about.

Why Japan.

For decades, Japan has been where industrial robots are built and put to work. The next generation of robots will learn from demonstrations, and Japan's factories are full of them.

38%
of the world's industrial robots are made in Japan.
IFR, 2025
446
robots per 10,000 manufacturing workers, more than three times the global average of 132.
IFR World Robotics 2025
10M
robots Japan aims to deploy by 2040 under its national AI robotics strategy.
Government of Japan, 2026

The robots are already on the line.

Japanese plants run industrial and collaborative robots every day. Recording teleoperation data fits into cells that already exist, with the parts, fixtures and lighting of real production.

The work is standardized and repeated.

Decades of monozukuri and kaizen have produced documented processes, clear success criteria and operators who perform a task the same way, shift after shift. That is what good demonstrations look like.

The need for automation is urgent.

With an aging, shrinking workforce, manufacturers have every reason to teach robots new skills. Sharing data with the teams building those skills is a natural next step.

Every listing answers the questions a training team asks.

What was recorded, on which robot, how it was labeled and under which terms. Buyers see it all before downloading a sample.

Example listing

Connector insertion, automotive wire harness line

Task
Inserting wire harness connectors into a housing
Collection
Leader–follower teleoperation by trained line operators
Robot
Dual 6-DoF arms with parallel grippers
Recorded
3 RGB cameras, joint states and actions at 30 Hz
Volume
120 hours, 4,800 episodes
Quality
Every episode labeled for success, failure and recovery
Rights
Operators consented; faces and part numbers masked
License
Commercial, non-exclusive. Exclusive on request
Format
LeRobot v2
See the datasets

How it works.

If you train robots

  1. 1

    Browse or ask

    Browse the datasets, or tell us the task, robot and volume you need and we will find it.

  2. 2

    Evaluate

    Download sample episodes and check what was recorded before you commit.

  3. 3

    License

    Airion handles the contract and payment. Download the full dataset once the license is granted.

If you run a factory

  1. 1

    Collect

    Record with PAC Build or your existing setup. We can help you design the collection.

  2. 2

    Prepare

    Mask faces and confidential details, label each episode, and upload a sample and the full dataset.

  3. 3

    Earn

    Airion checks the data and publishes your listing. Buyers receive the full dataset only after a deal.

Part of PAC, the Physical AI Canvas.

PAC Build makes Physical AI easy to build. PAC Exchange makes physical data easy to trade.

PAC Build

Connect robots, cameras and input devices. Record, label, curate, train and deploy, with quality checks on every episode.

PAC Exchange

Turn the data you already collect into revenue, or source the exact demonstrations your models are missing.

Questions.

Get in touch.

Looking for manipulation data you can't find anywhere else, or sitting on data that could train the next generation of robots? Talk to us.

Create an account