Scale Robotics

THE HUMAN LAYER FOR ROBOTICS

Not live yet. Beware of impersonators.

[Scale Robotics @ScaleRobotics]

Introducing Scale Robotics, a live platform turning human skill into robot intelligence.

Control humanoids in simulation, capture RGB and depth observations, record demonstrations, and generate training data for behavior cloning, Diffusion Policy, reinforcement learning, and VLA models like OpenVLA.

We are building the human data layer for physical AI, where thousands of people can teach robots how to operate in the real world.

This is how robot intelligence scales.

https://t.co/9p5m8eWKkr

[Scale Robotics @ScaleRobotics]

@Mrk_online Yes!

[Scale Robotics @ScaleRobotics]

Introducing Scale Robotics, a live platform turning human skill into robot intelligence.

Control humanoids in simulation, capture RGB and depth observations, record demonstrations, and generate training data for behavior cloning, Diffusion Policy, reinforcement learning, and VLA models like OpenVLA.

We are building the human data layer for physical AI, where thousands of people can teach robots how to operate in the real world.

This is how robot intelligence scales.

https://t.co/9p5m8eWKkr

[Scale Robotics @ScaleRobotics]

Humans already know how to open drawers, organize kitchens, move objects, recover from mistakes, and solve thousands of physical tasks.

Robots need to learn it.

Scale Robotics is building a way to transfer that knowledge.

A human controls a robot inside simulation. We capture observations, actions, trajectories, task outcomes, and corrections. Those demonstrations become training data for embodied AI.

Instead of manually programming every behavior, we can teach robots by showing them what to do.

Millions of people know skills robots will eventually need.

We want to turn that human experience into machine intelligence.

[Scale Robotics @ScaleRobotics]

We are working on a system where humans can teleoperate robots inside simulated environments while we capture synchronized observations, actions, trajectories and task outcomes as structured training data.

Currently the focus is on control latency, data quality, episode recording and making the whole loop reliable enough for real use.

More soon.

[Scale Robotics @ScaleRobotics]

We're building interactive robot teleoperation on MuJoCo physics and neural locomotion.

The goal is to make human demonstrations reusable, so teaching a robot a skill doesn’t mean programming it from scratch.