Buy My Robots
Back to home

Nov 21, 2026 · Buy My Robots Team

X-Humanoid says home humanoid robots start with laundry, not viral kitchen demos

El País published Jian Tang's November 21 interview on how X-Humanoid humanoid robots reach homes. Buyers should treat demo reels as site-specific data, not proof a used biped can run your kitchen.

Humanoid robots that cook on camera are still easier to sell than humanoid robots that survive a stranger's kitchen. On November 21, Spain's El País published a long interview with Jian Tang, chief technology officer of X-Humanoid, the commercial name for Beijing's Humanoid Robot Innovation Center. Tang's team also ran the Tien Kung Ultra bodies that grabbed sprint headlines at the World Humanoid Robot Games. Today's story is not another track record. It is a factory insider naming the gap between polished demos and the home humanoid SKU buyers will actually want on Buy My Robots.

What Tang said about demos versus real homes

Tang told El País the biggest public mistake is assuming a complex cooking clip means you can order a home humanoid next year. Demos work because operators collect huge amounts of data inside one fixed room. Drop the same policy into thousands of unfamiliar homes and the robot hits objects it never saw, then freezes or fails.

He used garlic in a fridge as a precision example. With thousands of recorded demonstrations in a known layout, cutting small food should be workable. In a new home, the same task is not repeatable yet. That is a direct warning for anyone pricing used humanoids off viral reels without task tables tied to floor plans.

Tang still believes home humanoids become family-scale tools. He just stretched the calendar. Within four years he expects true domestic service products, but limited: fetch a glass of water, simple handoffs, narrow chores. Within five years he pointed at laundry cycles with a washer, folding, and basic cleaning. Within nine years he said a unit might handle perhaps 90% of household tasks, while admitting success rates will not hit 100% on every attempt.

The first paid chores he named are boring on purpose: pick and place objects, including fragile cups and bottles, then load a washer, start it, unload, and fold. That sequence matters for listings because it is how engineers prioritize grasp policies, not how marketing teams order sizzle reels.

Price, software, and why teleop still sells

On cost, Tang said volume manufacturing should pull consumer humanoids toward roughly $20,000, comparable to a car once factories scale. That number is a target, not a quote sheet. It still sets a benchmark against 1X NEO deposits, Unitree R1 entry pricing, and industrial Digit or Figure contracts priced like capital equipment.

He blamed software, not leg motors, for the slow rollout. The sector is waiting for an embodied intelligence jump he compared to ChatGPT's generalization in text. Until a physical-world model handles novel cups, pens, and layouts, teleoperation stays the honest way to make humanoids look capable on video. Tang said remote control can do almost anything today; the hard part is repeating it autonomously after the operator logs off.

That mirrors Unitree founder Wang Xingxing's August timeline speech, but Tang's calendar is home-specific and task-ordered. Wang framed unfamiliar-home voice commands as a multi-year industry milestone at an investor-heavy conference. Tang, speaking to a European newspaper on November 21, attached dates to laundry, cleaning, and object search chores buyers can diligence.

Hardware choices, durability, and privacy

Tang defended bipeds for homes while admitting wheeled humanoids win inside factories where speed and precision beat stairs. For domestic buyers, he argued familiar human shape lowers psychological friction, comparing prior work on voice-controlled air conditioners that people refused to chat with for ten minutes.

He claimed current X-Humanoid bodies tolerate falls better than last year's hardware, citing internal drop tests where units stood back up across 100 trials, outcome dependent on how they land. He pointed at soft electronic skin overlays that stop motion on light contact, plus ambitions for five-finger hands that look human enough for living rooms.

Privacy came up directly. Tang said the company is pushing vision-heavy computation onto the robot body or a small home server so intimate video does not stream to the cloud by default. Operators reselling humanoids should ask where inference runs, what logs leave the building, and whether a new owner can wipe telemetry before transfer.

What this means on a humanoid marketplace

Sellers listing X-Humanoid, Unitree, AgiBot, or Western research platforms should separate three proof tiers Tang implied without using accounting language:

  • Stage or sports demos that stress stability algorithms, not transferable chores
  • Teleop or heavy data-collection sessions that make a single kitchen look solved
  • Narrow autonomous bundles with logged success on named tasks in named rooms

Buyers comparing Tesla Optimus, Figure 03, Boston Dynamics Atlas, and Agility Digit should apply the same filter. Factory and warehouse hours remain the liquid comps. Home generalization is still a forecast column, even when the CTO of a marathon-famous Chinese brand says the sector is "purely" a software problem.

Secondary pricing should punish ambiguity. A humanoid marketed as "home ready" without disclosure of teleop hours, demonstration map coverage, or laundry-task logs is carrying Tang's warning label in the fine print. A unit sold as a sports or exhibition platform should trade closer to entertainment robotics, not to 1X NEO preorder math.

Insurance and landlord rules will follow edge-compute claims. If Tang's on-device vision pitch becomes a shipping feature, resale packets need firmware notes showing whether cloud backup was disabled and whether edge models transfer with the sale.

What to watch next

Watch whether X-Humanoid publishes laundry or pick-and-place benchmarks outside Beijing demo flats before the four- and five-year windows Tang named. Watch whether European buyers at the China-Spain Green Energy Forum in Madrid, where Tang was scheduled to appear the week of November 21, translate his timeline into pilot contracts or stay at keynote level.

Western home programs still run on different clocks. Figure publishes zero-shot home task failure rates. Tesla keeps Optimus inside Fremont lines. 1X sells NEO with supervised expert mode. Tang's November 21 quotes do not reset those roadmaps. They do give marketplace users a second C-level calendar that ranks laundry ahead of gourmet cooking, and $20,000 ahead of mystery preorders.

Humanoid robots will keep dancing for cameras. The purchase decision, Tang argues, belongs to the boring first chores and the software generalization gap in between.

Sources