The question for anyone tracking home robots isn't whether Xiaomi can build a humanoid. It's whether the company that put a robot vacuum in millions of living rooms can do the same with a full-size humanoid — and when.
Here's what we know, what the data shows, and what it actually means for your home.
From CyberOne (2022) to CyberOne V2 (2026)
Xiaomi first unveiled CyberOne in August 2022. It stood 177 cm tall, weighed 52 kg, and could walk at 3.6 km/h. It recognized 45 human emotional states and could perceive its environment in 3D. By the standards of 2022, it was an impressive demo. But it was also clearly a research prototype — no price, no timeline, no real-world tasks beyond waving on stage.
Four years later, the V2 is a different machine. Here's what's changed:
Spec
Height
- CyberOne (2022)
- 177 cm
- CyberOne V2 (2026)
- 178 cm
Spec
Weight
- CyberOne (2022)
- 52 kg
- CyberOne V2 (2026)
- ~52 kg
Spec
Walking Speed
- CyberOne (2022)
- 3.6 km/h
- CyberOne V2 (2026)
- ~3.5 km/h (0.98 m/s)
Spec
Hand DOF
- CyberOne (2022)
- Not disclosed (basic grippers)
- CyberOne V2 (2026)
- 22–27 degrees of freedom
Spec
Single-Arm Lift
- CyberOne (2022)
- 1.5 kg
- CyberOne V2 (2026)
- 3 kg
Spec
Tactile Sensing
- CyberOne (2022)
- None
- CyberOne V2 (2026)
- Full-palm, 8,200 mm² coverage
Spec
Thermal Management
- CyberOne (2022)
- None
- CyberOne V2 (2026)
- Active liquid cooling ("sweat glands")
Spec
AI Model
- CyberOne (2022)
- Xiaomi AI platform
- CyberOne V2 (2026)
- Xiaomi-Robotics-0 (4.7B param VLA)
Spec
Status
- CyberOne (2022)
- Research prototype
- CyberOne V2 (2026)
- Factory-deployed, social demos
| Spec | CyberOne (2022) | CyberOne V2 (2026) |
|---|---|---|
| Height | 177 cm | 178 cm |
| Weight | 52 kg | ~52 kg |
| Walking Speed | 3.6 km/h | ~3.5 km/h (0.98 m/s) |
| Hand DOF | Not disclosed (basic grippers) | 22–27 degrees of freedom |
| Single-Arm Lift | 1.5 kg | 3 kg |
| Tactile Sensing | None | Full-palm, 8,200 mm² coverage |
| Thermal Management | None | Active liquid cooling ("sweat glands") |
| AI Model | Xiaomi AI platform | Xiaomi-Robotics-0 (4.7B param VLA) |
| Status | Research prototype | Factory-deployed, social demos |
The height and weight barely changed. Everything that matters — hands, brain, sensing — changed dramatically.
The Robot Hands That Sweat
The most genuinely novel hardware in the CyberOne V2 is its hands. Xiaomi Robotics Lab redesigned them from scratch, and the engineering tells you a lot about where humanoid robots are heading.
Human-scale proportions. The new hand is 60% smaller by volume than the previous version, designed to match a 1:1 human hand model based on a 173 cm adult. This matters more than you'd think — if the robot's hands are oversized, it can't use tools, door handles, or objects designed for humans. The sim-to-real transfer (training in simulation, deploying in reality) also works better when the robot's kinematics match human dimensions.
22–27 degrees of freedom. A human hand has roughly 27 degrees of freedom. Xiaomi's new hand gets to that range, up dramatically from the basic grippers on the original CyberOne. At Investor Day, the V2 units handed out gift bags, gave high-fives, and made heart gestures — movements that require coordinated finger articulation, not just open-and-close.
Full-palm tactile sensing. At 8,200 mm² of tactile coverage, the hand can sense pressure across the entire palm, not just the fingertips. This is critical for tasks like insertion and assembly, where the robot needs to feel whether a fastener has engaged correctly even when cameras can't see the contact point.
Bionic sweat glands. This is the headline feature, and it's real engineering, not marketing. High-density motors packed into a human-scale hand generate serious heat. Xiaomi integrated 3D-printed metal liquid cooling channels that circulate fluid through the hand — functionally analogous to how human sweat glands cool skin through evaporation. The result: the hand can sustain high-force grips for extended periods without performance degradation. Xiaomi reports the hand survived over 150,000 grasping cycles, compared to the roughly 10,000-cycle failure threshold common in tendon-driven robotic hands.
For comparison, see our analysis of why tactile sensing matters more than walking for home humanoid robots — the hands may determine whether a robot can actually do useful chores, not just walk around looking impressive.
Factory-Tested With Real Numbers
In March 2026, Xiaomi deployed CyberOne V2 units to its EV factory in Yizhuang, Beijing — the same facility that recently produced its 500,000th vehicle. The task: installing self-tapping nuts on an automotive assembly line.
The results were documented and shared publicly:
- 90.2% success rate for simultaneous dual-side nut installation
- 76-second cycle time matching the production line's fastest requirement
- 3 consecutive hours of autonomous operation
- Tasks included picking nuts from feeding devices, placing them on positioning fixtures, and coordinating with slide conveyors for automated tightening
This isn't a lab demo. A 90% success rate on a real production line with a 76-second cycle time means the robot is operating in the ballpark of usefulness — not ready to replace human workers, but no longer a science project either. Xiaomi is now testing the robot at additional workstations, including bin-picking and badge installation.
The Brain: Xiaomi-Robotics-0
The software side of CyberOne V2 is powered by Xiaomi-Robotics-0, a 4.7-billion-parameter Vision-Language-Action (VLA) model developed in-house. VLA models are becoming the standard "brain" architecture for general-purpose robots — they take in visual and language input and directly output motor commands, rather than requiring separate perception, planning, and control modules.
Key details about Xiaomi's approach:
- End-to-end data-driven control, combining the VLA model with reinforcement learning
- Multimodal input: vision, tactile feedback, and joint proprioception (the robot's sense of its own body position)
- Hybrid motion architecture: an optimization controller solving each iteration in under 1 millisecond for real-time responsiveness, paired with a reinforcement learning controller trained through hundreds of millions of simulated random perturbations
- Reduced reliance on teleoperation data, which matters for scaling — you can't teleoperate thousands of robots simultaneously
Xiaomi has also developed TacRefineNet, a tactile refinement framework designed to work alongside Xiaomi-Robotics-0 for dexterous manipulation tasks. According to Humanoids Daily, the dual-model stack (TacRefineNet + Xiaomi-Robotics-0) is a key part of the robot's hand-control pipeline — a sign the company is thinking about the full manipulation stack, not just the high-level VLA model.
The VLA model isn't unique to Xiaomi — AGIBOT uses a similar approach with Genie, and Physical Intelligence's π0.7 represents the state of the art. But Xiaomi has something those companies don't: a consumer electronics supply chain and millions of existing smart-home customers.
Investor Day: From Factory Worker to Event Staff
In late April 2026, Xiaomi brought the CyberOne V2 to its Investor Day — not to demonstrate factory tasks, but to interact with people. Two units handed out gift bags, shook hands, and made heart gestures to attendees.
The shift matters. Factory tasks are controlled and predictable. Social interaction in a crowded room is not. The "clean timing" and natural cadence of the robot's gestures suggested significant progress in real-time whole-body coordination — the kind of capability you'd need for a robot navigating a home with children, pets, and clutter.
Xiaomi hasn't released pricing or a commercial availability date for CyberOne V2. CEO Lei Jun has said that large-scale humanoid deployment in Xiaomi factories will happen within five years (approximately 2030–2031), and has described the factory floor as a "training ground" for eventual home assistants.
Why Does Xiaomi's Position Matter More Than Other Humanoid Startups?
The humanoid robot field is crowded in 2026. Tesla is building Optimus, Figure AI is deploying humanoids at BMW, 1X Technologies is taking preorders for NEO, Unitree is selling humanoids on AliExpress, and AGIBOT just shipped its 10,000th robot. So why does Xiaomi matter specifically?
They already make home products. This sounds obvious, but it's a real differentiator. Tesla makes cars. Figure makes nothing else. Unitree makes quadrupeds and fitness equipment. Xiaomi makes phones, tablets, TVs, air purifiers, robot vacuums, rice cookers, security cameras, smart speakers — an entire ecosystem of consumer devices that already live in homes. If anyone can integrate a humanoid into an existing smart-home platform at scale, it's Xiaomi.
They have their own factory to train in. Startups struggle to find pilot programs. Xiaomi has millions of square feet of factory floor in Beijing, producing real vehicles at real scale. This "captive customer" advantage lets them iterate hardware and software at production speed without negotiating access to someone else's facility. We've seen this pattern with AGIBOT's factory-to-consumer pipeline as well.
They understand consumer pricing. The original CyberDog quadruped launched at roughly $1,785 — aggressive but not absurd for a developer-focused robot. Xiaomi's supply chain expertise in driving down component costs across phones and EVs could translate to humanoid pricing that undercuts competitors building from scratch.
They're investing in the AI stack. Xiaomi recently hired Luo Fuli, a prominent researcher from AI startup DeepSeek, to lead its MiMo large model team — strengthening the AI foundation that underpins both consumer devices and robotics. They're building the software "brain" in parallel with the hardware body, which is the right approach — as our analysis of robot app stores and downloadable skills showed, the software ecosystem will determine which robots are actually useful.
The Honest Caveats
None of this means you'll have a Xiaomi humanoid in your kitchen next year. Several important reality checks:
- No price or timeline. CyberOne V2 has no announced price, no preorder page, and no consumer availability date. Lei Jun's five-year factory timeline means consumer availability is likely further out.
- 3 kg arm payload is low. A single-arm lift capacity of 3 kg limits real household usefulness. Other humanoids in this class offer higher payload ratings, and home tasks like carrying laundry baskets or moving furniture require more payload.
- Walking speed is modest. At 0.98 m/s, CyberOne V2 is essentially unchanged from the original CyberOne's 3.6 km/h (1 m/s) spec — and well behind Unitree's H1 at 3.3 m/s. Speed isn't everything, but it matters for covering multi-room homes.
- Factory success doesn't equal home readiness. A 90.2% success rate on a single, highly controlled assembly task is impressive. A home contains thousands of unique objects, layouts, and edge cases. As Stanford's research showed — robots succeed at only 12% of household tasks in current testing — the gap between factory and home is enormous.
- Chinese supply chain questions. With ongoing U.S. scrutiny of Chinese robots, a Xiaomi humanoid may face regulatory barriers in Western markets even if the hardware is ready.
What to Watch
If you're tracking Xiaomi's humanoid progress as a potential buyer or just an interested observer, here are the signals that matter:
- Expansion to more factory tasks. Xiaomi says it's testing bin-picking, badge installation, and other workstations. If the success rate holds across diverse tasks, that's a strong signal of general-purpose capability.
- CyberDog 3 or successor. Xiaomi's quadruped line is available for consumers (CyberDog 2 at ~$1,785, China-only). A next-generation quadruped with VLA capabilities would be a stepping stone to a consumer humanoid.
- Mi Smart Home integration announcements. If Xiaomi starts talking about CyberOne as part of its smart-home ecosystem rather than just a factory tool, the consumer timeline is accelerating.
- Pricing signals. Watch for any hint of a price range. Given Xiaomi's history of aggressive pricing in phones and EVs, a humanoid under $50,000 — or even under $20,000 — would be a market-defining move.
- International regulatory posture. How Xiaomi handles data privacy, on-device processing, and geographic server choices for its robots will determine whether it can sell in Western markets at all.
Bottom Line
Xiaomi's CyberOne V2 is one of the more interesting humanoid robots in 2026 — not because it's the most capable (Unitree and Figure currently have stronger specs in several areas), but because of the company behind it. A consumer electronics giant with its own factory training ground, an existing smart-home ecosystem, and a demonstrated willingness to price aggressively is a different kind of threat than a well-funded startup.
The bionic hands with "sweat glands" represent genuine engineering progress on one of the hardest problems in humanoid robotics — dexterous manipulation at human scale. The 90.2% factory success rate shows the robot can do real work, even if only one task so far. And the Investor Day demo proved it can function around people, not just on an isolated assembly line.
But "interesting" isn't the same as "ready for your home." Xiaomi is following the factory-first path that every major humanoid company has adopted, and the honest timeline for consumer availability is years away, not months. The company to watch? Yes. The robot to preorder? Not yet.
Track the CyberOne V2 and compare it against other humanoids in our home robot database, and see how it stacks up against competitors on our compare page.
Last updated May 4, 2026. Specs and claims sourced from Xiaomi official announcements, CNBC, CnEVPost, Interesting Engineering, Humanoids Daily, and 36Kr. Factory success rate is Xiaomi's own reported figure from controlled conditions and has not been independently verified.
Written by
ui44 Team
Published May 4, 2026
Share this article
Open a plain share link on X or Bluesky. No embeds, no widgets, no cookie baggage.