RESEARCH

Research

Our own factory is the R&D site. We publish the experiments we run on real dual-arm robots — the ones that worked and the ones that did not.日本語

HOW WE RESEARCH

How we do research

Bet on one industry

We are not building a robot that can do anything. The axis is not breadth of capability, but whether a linen factory can run it and make money.

Measure everything in yen per hour

A 90% success rate is not a goal by itself. Every research goal reduces to the profit one robot generates in one hour.

Evaluate results honestly

Controlled comparisons, and metrics designed before training. When we find a confound, we discount our own numbers ourselves.

ARTICLES

Articles

2026.08.17RECAPVLAManipulationReal-robot evaluation

Folding Towels, Measured in Yen per Hour

We put advantage-conditioned learning (RECAP) on a real dual-arm robot and cut towel-folding cycle time by 4 seconds (over 5%); success rate barely moved. Every result is measured in profit per robot-hour.

杉山幹太Head of ML EngineeringRead

JOIN THE TEAM

Come fold towels with us

The people who run these experiments also design the metrics and stand next to the robot in production. We are hiring ML engineers.

Request a casual chat See open positions