Unitree UniStore: Are Robot Skills Safe at Home?

Phone apps fail by freezing or draining a battery. A humanoid action can fail by falling, striking furniture, slipping on a floor, or surprising a person in the room. So the useful question is not simply "can my robot download new skills?" It is: what physical risk comes with the skill, and who checked that it fits my robot and my home?

The short version: UniStore is a real signal that robot abilities are becoming more software-like, but today's catalog is not a household chore store. In an official API snapshot read by ui44 on 2026-05-26, UniStore listed 27 actions: 16 dance actions and 11 martial-arts actions, all targeting G1 Edu or G1 Edu+. That makes it a good safety case study before future home robots start offering object handovers, tidying routines, kitchen skills, or elder-care behaviors.

What is Unitree UniStore?

UniStore is Unitree's action platform for downloadable robot behaviors. The public web page is sparse, but the official UniStore API exposes a live action catalog with names, categories, supported models, preview media, version numbers, publish times, download counts, and manual text for individual actions.

As of the snapshot used for this article, the catalog was narrow but meaningful:

That distinction matters. A dance or martial-arts routine is a bounded demo motion. It can still be risky, but it does not claim to understand your home, recognize fragile objects, or adapt to a pet crossing its path. A future "clear the coffee table" skill would have to combine navigation, perception, gripping, force limits, recovery behavior, and privacy controls. UniStore shows the storefront shape. It does not yet prove the consumer-home chore layer.

This is also why the existing broader question of robot app stores needs a second, more physical checklist. Downloading a behavior is only the easy part. Running it safely is the product.

Why the G1 front flip is the useful warning sign

The most instructive UniStore listing is not a household task. It is action id 185, named G1-韦伯斯特前空翻, a Webster front flip for G1 29-DOF hardware. The action description says the robot performs a complete 360-degree forward flip: standing start, crouch, jump, airborne rotation, landing buffer, and return to standing. Unitree's action info says it moves forward about 1.3 meters and reaches roughly a 1-meter airborne peak.

That is a remarkable downloadable motion. It is also a clear reminder that a robot skill can be a kinetic event, not a convenience feature.

The listing's own prerequisites are the headline:

• G1 29-DOF movable-waist hardware only, with mode-machine values 5, 11, 13, or 15.
• Not for 23-DOF versions, locked-waist versions, or dual-arm variants; the action exits on unsupported modes.
• Robot standing stably in motion-control state before launch.
• Battery recommended above 50%.
• A flat, hard floor that is not too slippery.
• Clear space of 2 meters in front, 2 meters left and right, and 1 meter behind.

Those requirements are not legal fine print. They are the core product truth. If a skill needs a hard floor, a specific waist configuration, more than half a battery, and a several-meter clear zone, a home user should not treat the install button like a toy button.

The official Unitree G1 product page already points in the same direction. G1 starts at $13,500, stands 132 cm tall, weighs about 35 kg, runs for about 2 hours, and can be configured from 23 to 43 degrees of freedom. Unitree's own page warns that humanoids have complex structures and powerful output, asks users to maintain sufficient safety distance, and advises individual buyers to understand humanoid limitations before purchase. That is unusually blunt, and buyers should take it seriously.

How Unitree compares with other skill ecosystems

Unitree is interesting because it is pushing software distribution onto capable, relatively affordable hardware. But different robots represent different risk profiles.

This comparison is not about picking a winner. It is about matching the skill to the body. A desktop robot can safely experiment with personality behaviors that would be irresponsible on a full-size humanoid. A quadruped can recover from some locomotion errors differently from a biped. A 70 kg humanoid needs a much stricter review layer than a small companion bot.

The market is moving toward skill libraries, but the safest route may be uneven: small robots get open community behaviors first, research robots get developer pipelines, and home humanoids get manufacturer-reviewed physical skills with strong permission gates.

What would a home-ready robot skill store need?

For household buyers, a robot app store should eventually look less like a phone store and more like a safety case. The install screen should answer seven questions before the first motion runs:

1. Which exact robot models and hardware revisions are supported?
2. What room size, floor type, lighting, and obstacle clearance are required?
3. Which actuators, sensors, cameras, microphones, and network services does it use?
4. What are the speed, force, payload, and balance limits?
5. Does a human need to supervise or hold a controller?
6. How does the robot stop, recover, and log failures?
7. Who is liable if the downloaded behavior causes damage or injury?

That list sounds heavy because the product is heavy. A robot skill is software plus mass, torque, batteries, perception, and judgment. Buyers should expect the same clarity they would expect from a power tool, not the same casual install flow they expect from a weather widget.

The Seoul Shinmun home-robot safety feature quoted Korean robotics experts making the same broader point: humanoids entering homes will need integrated safety approaches, objective certification, and guidance because children, elderly people, patients, and pets raise the stakes. The exact regulatory path will vary by country, but the buyer principle is universal: when the robot moves around people, software distribution becomes a safety system.