Donut Robotics Prepares Cinnamon 2 Humanoid for End-2026 Lease
Japan’s robotics industry is pushing “physical AI”—systems that sense, decide and act in the real world—as startups and manufacturers try to turn factory know-how into an edge against faster U.S. and Chinese rivals, News On Japan reported on September 28, 2026. Tokyo-based Donut Robotics is preparing Cinnamon 2, a humanoid that learns fluid motion from video or motion capture, follows spoken instructions, reads gestures, and targets security, construction, nursing care and factory tasks, with an official lease launch planned by the end of 2026 and a formal unveil due in October.
Filed under Industry and dated September 28, 2026, this AI4Japan briefing treats Donut’s Cinnamon 2 as Japanese physical-AI product news distinct from yesterday’s Sumitomo Life agent-tooling story. CEO Daisuke Ono said the vision-language stack is developed in-house while hardware is currently built in China under OEM arrangements and assembled in Japan, with further design changes planned for 2027; early commercial bets favour patrol and inspection before harder factory autonomy.
Why it matters: Japanese plants still sit on proprietary motion and safety data rivals cannot scrape from the open web. Physical AI can monetise that edge—but only if human stop authority, identity for machines and workplace safety cases travel with every lease.
What it means in practice
Japanese manufacturing and facilities leads should inventory which patrol or inspection routes could host supervised humanoids; demand named safety owners and lease kill criteria; assign an owner for machine-ID and data-sharing agreements; run time-boxed pilots against wheeled baselines; and prefer designs that keep humans on irreversible actions. Anchor Cinnamon 2 to Japan’s AI-robot labour push and factory machine-ID data plans.
Caveats come first. An end-2026 lease target is not a nationwide fleet; OEM supply chains can shift; and gesture demos are not certified care robots. AI4Japan therefore presents Cinnamon 2 as directional industry context until published deployment metrics appear.
What to watch next: October unveil details; first security-customer cohorts; and whether government machine-ID schemes feed Donut’s training loop. Readers can continue on the AI4Japan homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Japanese physical AI is pairing humanoid leases with factory data strategy and remains early. Organizations that benefit most will demand safety evidence, keep humans on stop authority, and refuse to confuse a demo with finished autonomy.