Research
Our methods, our test rig and, soon, our results.
We publish how we test before we publish any scores, so anyone can question our choices. Every report will come with the runs, videos and data behind it. Our first results are not out yet.
Methods
Genba Bench: how we test everyday factory work The tasks, scoring rules and setup for our capability benchmark. Anzen Bench: how we test refusal of unsafe instructions How we write unsafe instructions in Japanese and English, and how we score a refusal.Behind the scenes
Our Tokyo test rig The arms, cameras and controllers we use. Embedded devices from Renesas.News and essays
JETRO, 99VC, AWS Startups and NVIDIA Inception Why Japan needs a public scoreboard for robot AIReports
No reports published yet. Our first Genba Bench report is in progress.
