China has crossed a psychological and industrial threshold: a dedicated factory in Liuzhou is now set up to build humanoid robots at a 10,000‑unit annual scale, with robots helping to move parts and manage flow—an early glimpse of “robots building robots” shifting from stage demo to configured production line.
At a Glance
- UBTECH has commissioned a purpose-built humanoid robot plant in Liuzhou, Guangxi, designed for 10,000+ units per year.
- The site integrates digital manufacturing, automated intralogistics, and mixed-model assembly for Walker S and Cruzr platforms.
- Officials and company statements describe a takt of one industrial humanoid robot every 10 minutes at steady cadence.
- This milestone aligns with China’s broader strategy to extend its industrial-robot dominance into humanoids at scale.
What was commissioned in Liuzhou—and why it matters
UBTECH’s Liuzhou “industrial humanoid robot super smart factory” has been publicly introduced as a mass-production facility for the company’s Walker S and Cruzr product lines, with a design target above 10,000 units per year. Reports describe a roughly 14,000‑square‑meter, 13.8‑meter‑tall building configured for mixed‑model humanoid assembly, end‑of‑line testing, warehousing, and shipping under a single digital management layer. The company frames the site as moving beyond prototype showmanship into repeatable throughput—a core requirement for driving costs down and reliability up in a category long stuck in pilot projects and trade‑show theatrics.
The commissioning ceremony and subsequent coverage emphasize an integrated “closed intralogistics loop” in which automated guided vehicles (AGVs) and coordinated software orchestrate materials flow, with robots supporting portions of production. The stated steady‑state takt—one robot approximately every 10 minutes—underscores the ambition: a shift from bespoke assembly to cadence manufacturing, where predictable station timing and quality checks enable output scaling without crippling rework downstream.
How a humanoid factory works: from modules to mixed-model lines
Humanoid production is less like smartphone assembly and more like building compact electromechanical athletes. The line decomposes robots into standardized subassemblies—actuated joints, limbs, torso, compute and power modules—each with torque‑dense motors, harmonic drives, battery packs, sensor suites (IMUs, depth and radar where used), and thermal and wiring harnesses. A digital backbone ties procurement, traceability, and test results to each serial number. Mixed‑model operation means Walker S variants and service‑oriented Cruzr units can traverse shared workcells with recipe changes, fixturing swaps, and software provisioning adapted on the fly. Automated logistics—AGVs shuttling totes and frames—reduces non‑value‑added time and enforces first‑in, first‑out discipline at buffers, while end‑of‑line stations run gait, balance, manipulation, and safety checks before warehousing and outbound booking.
None of this is exotic in principle—automotive and electronics plants have run digital twins and closed loops for years—but applying it to bipedal systems with high degrees of freedom and stringent fall‑safety requirements is a real engineering lift. A humanoid’s integration burden runs through mechanics, power electronics, embedded control, and embodied AI; productionizing that stack demands standardized interfaces, robust calibration routines, and a test regime that catches intermittent faults (e.g., marginal encoders or thermal derates) before field deployment.
What “one every 10 minutes” implies in practice
A takt claim of 10 minutes is a statement about station design, not a guarantee of calendar output. Still, it’s meaningful. Assume a 20‑station line balanced to 10 minutes per station; a single shift would push dozens of units across end‑of‑line, contingent on upstream kitting and yield. The more critical metrics are first‑pass yield, mean time to repair for station faults, machine utilization, and percent of labor content performed autonomously versus by skilled technicians. The Liuzhou plant’s positioning suggests UBTECH has engineered for line cadence, digital traceability, and intralogistics maturity—the triad necessary to convert capital equipment into repeatable shipments.
The products tied to the site matter too. Walker S targets industrial humanoid roles—higher payload and precision for factory‑adjacent tasks—while Cruzr leans toward service and reception scenarios. Building both under one roof forces process generality: fixturing that tolerates height and mass differences, software flashing and calibration that handle variant SKUs, and test scripts that gate capabilities appropriately before release.
Why Liuzhou—and why now
Industrial policy and supply‑chain density set the table. China leads the world in installed industrial robots, with more than two million operating in factories by mid‑decade; that base is supported by local component ecosystems that compress sourcing time and cost. Shenzhen, in particular, can deliver motors, geartrains, sensors, and control boards in days rather than weeks, a structural advantage when iterating fixtures and subassemblies for a new line. The Liuzhou buildout is being presented as a municipal and regional milestone: a benchmark facility signaling that Guangxi is competing for the next phase of embodied‑AI manufacturing, not just classical industrial automation.
Globally, this is part of a visible push by Chinese firms to scale humanoid output—from reception and logistics helpers to factory‑floor assistants—and to convert R&D showcases into revenue programs. Multiple reports and factory tours across Shanghai, Guangdong, and Beijing have documented lines purpose‑built for thousands of humanoids annually. The Liuzhou site sits squarely in that arc: an explicit attempt to industrialize the category rather than celebrate prototypes.
What to watch next: from ramp to reliability
Commissioning a line is step one; sustaining it is the game. Over the coming quarters, the meaningful signals will be deployment mix (industrial versus service use cases), field reliability (failure modes, maintenance cycles), and learning velocity as fleets generate the data that improves manipulation and autonomy. China’s manufacturing playbook—rapid iteration, local supplier co‑development, and capital intensity aligned with policy goals—positions plants like Liuzhou to move fast once early defects are burned down. And if the takt promise holds under real utilization, unit costs will follow the familiar curve: higher volumes enabling better components, tighter process control, and then more capable robots at lower price points.
Sources:
insiderpaper.com, pandaily.com, liuzhou.gov.cn, en.people.cn, x.com, interestingengineering.com, kantan.news