Inside a Chinese garment workshop, stacks of cut fabric pieces wait to be fed into template-sewing machines. The task looks trivial to a human — pick up a soft panel, place it squarely, align the edge — but it has resisted automation for years because cloth folds, curls, and shifts unpredictably. At one station now, a humanoid robot (人形机器人) does exactly that: separating pieces, grasping flexibly, laying them flat, and aligning them with precision, hour after hour.
This is not a trade-show demo. It is the Aitu (艾图) humanoid, built by Jack Technology (杰克科技), the global leader in intelligent sewing equipment, and already running on real production lines in several domestic clothing factories.
Why clothing was supposed to be impossible
Soft, irregular fabric is close to the worst case for a robot. Hard to locate, easy to deform, impossible to "fixture" the way you would a metal part. That is precisely why garment manufacturing stayed stubbornly manual even as car and electronics plants automated. Aitu targets the repetitive, low-variability steps first: loading and unloading template machines, placing fabric for pocket-attaching, and handling cut pieces at fusing machines.
The company reports that under single-category, single-fabric conditions, Aitu's piece-separation success rate reaches 97% and its sewing qualification rate exceeds 98%. On that efficiency, the payback period for one unit is estimated at about 18 months. Those numbers are narrow — they assume a stable product and material — but they are the kind of figure a factory manager can actually model.
The strategic stakes are larger than one workshop. China is the world's largest garment producer, and the industry employs millions in precisely the repetitive, high-turnover stations that robots now target. A humanoid that can reliably handle soft fabric would unlock a far bigger addressable base than the car and electronics lines that dominate today's humanoid headlines. That is why a sewing-equipment maker, rather than a general-purpose robot startup, may be the right company to crack the problem: it already understands the garment process, the machines, and the ergonomics of the floor.
From demo to deployment
Several production lines are now in pilot mass-run mode rather than one-off trials. In April 2026, Aitu partnered with the Zhejiang Humanoid Robot Innovation Center (浙江人形机器人创新中心), Zibian (自变量), and AgiBot (智元) to push a systematic upgrade of garment manufacturing — its supply chain, its production model, and its value network. The bet is that real factory data, fed back into the model, will widen those 97–98% rates toward industrial-grade consistency.
What "entering the factory" really means
The garment case fits a broader pattern in China's embodied AI (具身智能) push: robots are being put first into single, well-bounded, repetitive tasks where failure is tolerable and the return is calculable. Industry observers note that the robots now doing real work share four conditions — clear task boundaries, single repetitive motion, a measurable return-on-investment model, and a job that is hard to staff or physically taxing.
Crucially, this is a structural supplement, not a replacement. As one executive put it, the annual natural retirement and attrition of garment workers far exceeds any number of robots deployed. "Recruitment difficulty" (招工难) is the real driver; the robot fills a gap rather than eliminating a role.
Reading the trend
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the fastest path to useful humanoids is letting the production line itself become the training ground, because real-work data improves the model far faster than a laboratory ever can.
The policy tailwind is real, too. The Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission launched a "real-scene training" (实景实训) initiative for humanoids and embodied AI, aiming to switch on "work mode" by the end of 2026 and build ten-thousand-unit-scale deployment capacity. Local governments are adding purchase subsidies for domestic embodied equipment.
Honest limitations
The 97% and 98% rates and the 18-month payback are company-disclosed and tied to a single product type and fabric; they are not independently audited and will degrade as material variety rises. "Several factories" and "pilot mass-run" are vague deployment descriptions, not installed-base figures. Generalizing from flat fabric panels to the full chaos of a cutting room is unsolved. Treat this as an early, promising foothold — not evidence that humanoids are about to run entire apparel lines.
What readers can do now
- For manufacturers: if you run repetitive, hard-to-staff stations (not just garments — electronics, auto parts), request a bounded pilot with explicit success-rate and payback reporting before scaling.
- For investors: judge humanoid makers by deployed, paid, repeatable stations and the data flywheel they create — not by dance videos.
- For workers and the curious: the realistic near-term story is robots taking the dull, strenuous, high-turnover tasks while people move to oversight, maintenance, and process tuning. That transition is already underway on some Chinese sewing lines.