Mission

Dexterous manipulation infrastructure.

The single biggest bottleneck in robotics is the hand, and the high-quality data to train it. We build the hand together with the data engine that captures human work and the data that feeds the foundation model behind a robot that performs on a real production line.

01

Precise

Sub-millimeter accuracy and force-controlled contact, so the hand can take on the tight-tolerance, fragile-part work fixed automation can't.

02

Fast

Production cycle times, not research demos. Every layer of the stack is designed to hit takt time on a real line.

03

Reliable

Built for 24/7 uptime. Calibration that holds shift after shift, sensors that survive the factory floor, parts you can swap in minutes.

04

Low cost

Engineered for production economics from day one. The price point that makes general-purpose deployment make sense at every cell, not just every plant.

The problem

Two bottlenecks. And they're coupled.

Dexterous robots aren't blocked on one hard problem. They're blocked on two, and solving either one on its own gets you nothing deployable.

Bottleneck 01

Hardware that survives the line

As of 2026 there is still no reliable, easy-to-use, high-DOF dexterous hand you can buy off the shelf. Research hands don't hold calibration through a shift. Industrial grippers hold up fine but can't do the work. A production hand needs sub-millimeter precision and a 3M+ cycle lifetime, and almost nothing clears both bars.

Bottleneck 02

Data that actually transfers

Robotic foundation models need manipulation data at something like internet scale. It doesn't exist. Human video helps with pre-training, but the actions in it are inferred and noisy, and it carries embodiment, sensory, and control gaps from any real robot. Pre-training absorbs that noise. Deployment doesn't.

Here's why they can't be attacked separately. Robot data is instantiated by a specific body, and data from a different embodiment loses most of its value during post-training. So the data has to come from hardware with zero embodiment gap to the robot you actually deploy. Whoever solves this has to build the hand and the data engine as one system, not two products that meet at an integration milestone.

Team

A founding team with decades of experience delivering world‑changing technologies.

Decades across AI and robotics, tracing the same arc the field itself has taken: computer vision, then self-driving cars, now dexterous manipulation. We have built at production scale at every step of it.

Evan Tao

Co-founder & CEO

Tesla, Apple, and ASML. Shipped high-precision hardware from consumer devices at hundreds of millions of units to the lithography systems that print the world's most advanced chips — at volume, on cost, and to tolerance.

Joe Dong

Co-founder & CTO

Waymo and Xpeng. A decade in computer vision and autonomy, building the perception and learning stacks behind vehicles that drive themselves on public roads — where the long tail, not the demo, decides whether it works.

Where the team has been

Build with us.

We work with customers, investors, researchers, and engineers building toward the same goal.

Get in touch →