Cinnamon Mini: Small Humanoid Reality Check

That still makes Cinnamon Mini useful for home-robot buyers to understand. A small humanoid may reach useful reliability earlier than a full-size helper because it can be cheaper, lighter, easier to supervise, and less dangerous when it makes a mistake. But those advantages only matter if the robot proves more than choreography.

The short version: Cinnamon Mini is a credible signal that Japanese humanoid robotics may split into two tracks. Full-size humanoids chase physical labor. Small humanoids chase presence, guidance, interaction, entertainment, education, and supervised public-space jobs. For many households, that second track may be more realistic sooner.

What did Donut Robotics announce about Cinnamon Mini?

Donut Robotics' official April and May posts say Cinnamon Mini was first shown around SusHi Tech Tokyo 2026, the innovation conference held at Tokyo Big Sight from April 27 to 29. The April announcement said Cinnamon Mini would be unveiled publicly and appear in a live dance collaboration with Atarashii Gakko on the final day. The May follow-up summarized the same point more simply: Cinnamon Mini was announced at SusHi Tech 2026.

The more detailed official release adds the important buyer-facing facts. Donut positions Cinnamon Mini as the follow-up to Cinnamon 1, but with a different job. Where Cinnamon 1 is a roughly 170 cm, 70 kg full-size humanoid framed around factory and construction-site labor, Cinnamon Mini is described as a 130 cm small humanoid that is more lightweight, more agile, lower cost, and focused on entertainment use.

That framing matters. Donut is not saying Cinnamon Mini is a kitchen robot, a laundry robot, or an eldercare assistant you can buy next month. The official language points to customer service, staging, events, store traffic guidance, and dance-based spatial performance. It is a public interaction robot first.

The release also says Cinnamon Mini uses "motion learning from video" rather than relying on conventional motion capture. If that becomes robust, it could lower the cost of teaching humanoids new motions. For now, treat it as a promising development claim, not proof that a small humanoid can learn arbitrary household chores from YouTube.

Why a 130 cm humanoid could be more practical than a full-size one

A smaller humanoid gives up reach and payload. It also avoids some of the scary parts of putting an adult-size biped in a normal home.

A full-size humanoid like 1X NEO is closer to household work surfaces. ui44 tracks NEO at 167 cm, 30 kg, about 4 hours of battery life, and a $20,000 early-adopter price. That height makes sense for counters, appliances, shelves, and human-scale chores. But it also increases the safety burden: the robot has to move around people, pets, stairs, rugs, glass, chairs, and narrow hallways without causing physical harm.

By contrast, the existing compact-humanoid field is already crowded with robots that are interesting but limited. Unitree R1 starts around $4,900, stands 123 cm, weighs about 29 kg, and has about 1 hour of mixed-activity battery life. Unitree G1 is 132 cm and 35 kg, with a starting price of $13,500 and about 2 hours of battery life. NEURA 4NE-1 Mini is also 132 cm, 36 kg, from €19,999, and has about 2.5 hours of battery life with a listed 3 kg payload.

Cinnamon Mini sits conceptually in that compact group, but Donut has not yet published the key numbers that would make a real comparison possible: weight, runtime, payload, hand design, sensor package, price, shipping region, warranty, or buyer availability. So the honest read is simple. The 130 cm size claim is meaningful. The rest is still mostly an event-demo story.

The job may be performance before chores

The most interesting thing about Cinnamon Mini is that Donut is not pretending that every humanoid has to start by doing dishes. The official release talks about entertainment, event appearances, traffic guidance inside stores, and stage performance. That is a narrow lane, but it may be a sensible lane.

Robots fail less when the environment is controlled. A stage, shop entrance, event booth, museum lobby, school demo, or retail queue is easier to structure than an apartment. The floor can be cleared. The lighting can be controlled. The script can be rehearsed. A human can supervise. The robot can be useful even if it only handles a few repeatable behaviors.

That pattern already shows up in ui44's database. Pepper, the 120 cm social humanoid, survived in public and semi-public jobs because greeting, explaining, translating, entertaining, and scripted interaction are easier to package than general-purpose home labor. LOVOT, at 43 cm and 4.6 kg, succeeds in a different way: it is openly a companion, not a chore robot. SwitchBot KATA Friends is an available $699 AI pet companion, and Loona is a $429 small wheeled companion with face recognition, voice commands, games, and auto-docking.

Those robots are not substitutes for a humanoid helper. They are evidence that homes often accept robots when the promise is clear. Companionship is clear. Education is clear. Entertainment is clear. Store greeting is clear. "Maybe a humanoid can do anything" is not clear.

Cinnamon Mini's best near-term path may be to become extremely good at a few public-facing behaviors: moving with performers, welcoming visitors, guiding people through a space, responding to gestures, presenting information, and being safe around crowds. If it does that reliably, the robot can teach buyers something useful about compact humanoids even before it touches laundry.

Video learning is the claim to watch

Donut's most ambitious Cinnamon Mini claim is not the height. It is the statement that the robot can learn motion from video instead of conventional motion capture. In theory, that matters because motion capture is expensive, staged, and hard to scale. Video is everywhere. If a humanoid can turn video into safe, repeatable movement, the training pipeline gets much cheaper.

But buyers should separate three levels of evidence.

Level one is choreography. A robot copies a dance, pose, or rehearsed motion sequence. That is impressive, especially for a biped, but it does not prove household usefulness.

Level two is robust motion transfer. The robot can learn many motions from ordinary video, retarget them to its body, and repeat them under slightly different lighting, camera angles, and floor conditions. That would be valuable for entertainment and education.

Level three is task learning. The robot watches a human perform a practical job, understands the objects, adapts the motion to its own body, detects when the attempt is failing, and stops safely. That is the level home buyers actually care about. It is also the level that current public Cinnamon Mini evidence does not yet prove.

The distinction matters because the same phrase, "learns from video," can mean very different things. A dance routine is not a dishwasher routine. A public stage is not a cluttered kitchen. A scripted event appearance is not autonomous help in a home where a chair moved, a pet is underfoot, and a child interrupts the robot.

Cinnamon Mini vs. desk robots and mobile manipulators

For a buyer, the useful comparison is not only Cinnamon Mini versus full-size humanoids. It is Cinnamon Mini versus every smaller robot already trying to earn a place indoors.

That last row is the uncomfortable comparison for every small humanoid. If a buyer wants real physical help, Stretch 4 is ugly in the way serious tools are ugly: wheeled base, telescoping arm, sensors, open software, and a clear assistive/research role. A small biped has to prove that its human shape creates enough value to offset instability, lower payload, and harder control.

Cinnamon Mini may not need to win that comparison. It may be competing more with Pepper, KATA Friends, Loona, Reachy Mini, and event robots than with Stretch 4 or 1X NEO. That is a different market and a more believable first step.

What home buyers should ask before caring

Cinnamon Mini is worth following, but not because it is a near-term home purchase. It is worth following because it may show whether compact humanoids can become safe, expressive, and trainable enough to leave the lab.

Before treating it as a real home-robot candidate, wait for these specifics:

• Price and availability: whether Donut plans public sales, enterprise-only deployments, rentals, or event bookings.
• Runtime and charging: whether it can work through a normal event session without constant battery management.
• Weight and fall behavior: a 130 cm robot can still hurt people if it falls badly.
• Hands and payload: whether the robot can carry, point, gesture, grip, or only perform body motions.
• Sensor and safety stack: cameras, depth sensing, force limits, emergency stop, crowd behavior, and remote supervision.
• Video-learning proof: many learned motions, not one rehearsed routine.
• Privacy model: what cameras record, what goes to cloud AI, and whether a human operator can connect during demos or deployments.

The privacy question is easy to overlook because Cinnamon Mini is being shown as an entertainer. It should not be overlooked. A robot that moves through stores, venues, schools, or homes with cameras and microphones needs a plain data policy.

Bottom line: small humanoid, smaller promise, better test

Cinnamon Mini should not be read as proof that a small Japanese humanoid can beat full-size home robots at chores. It is more interesting than that. It is a test of whether a compact humanoid can pick a narrower promise and execute it better: perform, guide, interact, learn motions cheaply, and operate safely in supervised human spaces.

That is not the dream version of home robotics. It is not a robot butler. It is not a general housekeeper. But it may be a more realistic bridge than another full-size humanoid demo promising everything.

For now, the right verdict is cautious optimism. Cinnamon Mini has a sensible size, a clear event-first use case, and a genuinely important training claim in video-based motion learning. It still needs the boring spec sheet, safety data, runtime, price, buyer model, and repeated real-world demos before it belongs in a home-robot buying shortlist.