Humanoid Robot Hiring in China: Pay Bands and Gaps | SunTzu China

Humanoid Robotics Talent Research · Mass Production Series

A humanoid robot production line in Liuzhou now completes one robot every 10 minutes; Qiyuan runs 2.5 minutes per unit. The hiring centre of gravity is shifting from the lab to the floor.

Published by SunTzu China | Data as of 14 October 2026

Source tiers: government documents and official statistics / listed-company disclosures / market research reports / campus announcements and live postings

Coverage: complete machines, core components, supply chain and quality systems

Research question: humanoid robots have moved from expo demos into plants that ship in the tens of thousands. Has hiring moved with them? Public pay anchors still hang on R&D titles while the new headcount sits on the factory floor. This report uses no institutional talent-gap estimate as a denominator: it reads demand from takt and shipments, rebuilds two pay curves from campus and posting data, and splits the market by degree and city to locate the gap.

Key readings: global shipments up 432.1% year on year | a takt of 10 minutes per robot in Liuzhou and 2.5 minutes at Qiyuan | average payroll cost of RMB 358,700 per employee per year.

Robot quality engineer salary anchor: RMB 8,000-14,000 per month at campus level, against RMB 20,000-50,000 for the R& D track in the same graduating cohort.

First, shipments have crossed the ten-thousand-unit threshold, and the question has shifted from whether a robot can be built to whether it can be built consistently. IDC counts global first-half 2026 shipments near 25,000 units, up 432.1% year on year, China about 19,000 or 77.9%. Frost & Sullivan counts 36,000 units and USD 540 million of revenue, above full-year 2025. Omdia counted over 140 Chinese machine makers in 2025; only a minority manufacture at scale, and the top five held 73%.

Second, within one cohort the manufacturing track and the R&D track are two price curves. Unitree's 2027 campus hiring lists technical roles at RMB 20,000-50,000 a month; a peer head offers campus quality engineers RMB 8,000-14,000; a process engineer with two years is back at RMB 20,000-30,000 on a 13-month basis. Both enter at bachelor level, but manufacturing prices experience and systems capability while R&D prices degree and research direction, and the slopes point opposite ways.

Third, average payroll cost of RMB 358,700 per employee is the budget baseline of the production era. UBTech's 2025 payroll was RMB 914.6 million across 2,550 staff at year end; R&D staff numbered 1,103, about 43%, and 583 hold equity grants. The blended figure shows a budget still configured R&D-first even as the job mix tilts toward the floor.

Fourth, the manufacturing track's supply pool sits in automotive and consumer electronics, not inside robotics. Qiyuan organises production under IATF16949; UBTech's Liuzhou site runs a robots-building-robots loop. People who can hold batch consistency, yield and supplier quality at the ten-thousand-unit level sit in car plants, EMS and precision manufacturing, and cross-industry search is the main hiring path of this phase.

Data window: September 2025 to 14 October 2026 | Sample: three market research reports, one listed-company disclosure, two 2027 campus announcements, a set of live postings, two government documents and local disclosures | Excluded from evidence: salary breakdowns from aggregator Q&A pages, self-media pay lists, training-vendor job boards | Nature: observational, not sampled; conclusions cover the sample only.

I. Industry Landscape and Pay Bands

The policy floor comes first. The Ministry of Industry and Information Technology's 2023 humanoid robot guidance sets two milestones: by 2025 an initial innovation system, mass production and two to three ecosystem companies of global influence; by 2027 a secure supply chain at the world's advanced level. Official statistics add that China has taken over half of new global industrial robot installations in three years, with density at 470 units per 10,000 workers, up nineteen-fold in a decade. Policy milestones and capacity milestones began to coincide in 2026.

The demand side needs a scope warning up front. IDC counts about 25,000 units shipped globally in the first half of 2026; Frost & Sullivan counts 36,000 and USD 540 million, with the revenue top five being UBTech, AgiBot, Unitree, Galbot and Leju; a third house counts about 19,100 units, over 97% made in China. The scopes differ, so the report takes only the directional conclusion, that volume moved from thousands to tens of thousands within a year, and never mixes the three.

UBTech's 2025 annual results give the most complete headcount ledger of any whole-machine maker: revenue of RMB 2.001 billion, up 53.3%, the highest among global humanoid robot companies, including RMB 820 million from full-size industrial humanoid robots on 1,079 units; net loss of RMB 790 million, narrowed 31.9%.

At the end of 2025 it employed 2,550 full-time staff, 2,545 in mainland China, concentrated in Shenzhen with the rest in Liuzhou, Beijing and Xiamen. R&D staff were 1,103, about 43%, including 344 at master's level or above; 583 hold equity incentives; R&D expense was RMB 507.5 million. Static payroll of RMB 914.6 million implies about RMB 358,700 per employee per year.

The ledger also holds a capacity timeline. June 2025: the first Guangxi-made industrial humanoid comes off the line in Liuzhou. December: the thousandth Walker S2. 12 September 2026: the super smart factory built with Siemens starts production, 14,000 square metres, designed takt of one robot every 10 minutes, planned capacity above ten thousand units, robots taking part in assembly and logistics. In the same period the FAW-Volkswagen cooperation advanced from quality inspection to in-plant logistics. First unit, thousandth unit, ten-thousand-unit plant: three steps in fifteen months, and hiring's centre of gravity moved from lab to line.

A second timeline comes from Qiyuan Robot, a subsidiary of Shangwei New Material: automotive takt and quality management imported outright, one robot every 2.5 minutes, production under IATF16949. Its management's framing is worth quoting: building a single prototype is easy, holding batch consistency at the ten-thousand-unit level is hard, and the real challenge lies in the supply chain and manufacturing standards. AgiBot shipped more than 8,600 units in the first half, tenfold up. The question has moved from who can build one to who can ship, deliver and keep deploying.

The price curve completes the picture. Unitree's bipedal R1 fell from RMB 39,900 to 29,900; Songyan Dynamics pushed a bipedal product to RMB 9,998; Qiyuan's personal robots start at RMB 19,999. Goldman Sachs projects the average machine price falling from USD 41,800 in 2025 to USD 21,300 in 2035, nearly 49% in a decade. Falling prices rest on supply chain localisation and line efficiency, and efficiency is decided by the manufacturing track. Takt, shipments and price together give the shape of the problem: the prototype phase lacks engineers who can make a robot; the ten-thousand-unit phase lacks people who can hold consistency and yield.

Beneath the machines sits an employment base that aggregate statistics skip. Precision reducers, servo motors, encoders and joint modules form the cost and quality base, and the consensus at the 2026 World Robot Conference was that integrated actuator modules are consolidating discrete parts into standard units, simplifying assembly. Two consequences follow: process, inspection and equipment roles at component makers expand in step with machine volume, along a Suzhou, Dongguan and Changzhou belt with radius dependence on the assemblers; and the assemblers' own manufacturing profiles now include module assembly, calibration and burn-in testing.

The actuator and joint module supply chain is where that base sits, and it prices its own people: module integration and calibration roles now post alongside the assemblers' classic process and inspection lines. Counting only the assemblers understates the market's hiring demand by a third or more.

The table below consolidates the verifiable pay anchors by job family, as monthly bands with band centres. Two cautions first: campus figures are self-reported, live postings carry the platform's own basis, all are monthly, and none include equity or variable pay. The centres are computed from the bands and are not percentile statistics.

Manufacturing engineer salary China anchors remain thin: two campus announcements and two live postings is the whole verifiable record, which is itself a finding about how young this market's pay disclosure is.

Table 1: monthly pay anchors for key roles in the mass-production phase. Units are RMB per month except the two control rows, which are annual; compare within columns only. Band centres are computed from published ranges and do not constitute percentile statistics. The process engineer row is a 13-month basis, as marked. Data window: April to October 2026. Sources: Unitree and Dahan Robotics 2027 campus announcements (via university career channels), UBTech postings and annual results, live postings. For reference only, RMB 7.1 is treated as roughly USD 1.

The spread between job families is larger than the spread between companies. Within the same 2027 cohort and the same bachelor entry, the centre of the quality engineer band is RMB 11,000 while the top of the technical band reaches RMB 50,000, a gap of more than three times. Which family a candidate applies to moves the starting salary more than which company.

The two curves price on different variables. The R&D track prices degree and research direction, its campus band spanning a factor of about 2.5 from bachelor entry to doctoral top; the manufacturing track prices experience and systems capability, and a process engineer with two years lands at RMB 20,000-30,000, above most campus manufacturing offers. For candidates the curves describe two careers; for employers, manufacturing profiles must be written around systems capability rather than copied from R&D degree gates.

The all-staff figure is the budget anchor of the phase. UBTech's RMB 358,700 per employee sits below its own R&D campus and lateral technical offers and above its campus manufacturing prints. That middle position is what a production-era payroll looks like: a workforce that is 43% R&D takes the larger share of budget, and the new money that volume requires has to be found between diluting that share and lifting average cost.

Three steps make the curves readable. Step one, split R&D from manufacturing before comparing anything: one side reads on degree and research direction, the other on experience and systems capability. Step two, read past the print: campus figures exclude bonus, equity and shift allowances, and line roles often carry shift and project pay that never reaches the advertised number. Step three, read the human-to-robot ratio: lines this automated need fewer manufacturing roles per unit than traditional manufacturing, value per role is higher, and bargaining power concentrates in the people who carry systems capability.

II. Degrees, Experience, Cities and the Trends of the Production Era

The first section split the market by job family; this one splits it by the person. The structural change of the production era is a fork in the robotics career path: the R&D line still climbs the degree ladder, the manufacturing line climbs on systems capability and project record, and the two lines carry separate logics of appraisal, pay and mobility.

The R&D line runs on the degree ladder. Unitree's 200 campus roles cover all degree levels with technical pay at RMB 20,000-50,000, the top held by algorithm and control directions; head companies bind research staff with equity, and 583 of UBTech's 1,103 R&D staff hold grants, more than half. Entry gates on degree and project record; output is measured in patents, models and iteration speed.

The manufacturing line runs on systems capability. A live posting for a humanoid line process engineer asks for assembly process design, dimension chains and GD&T, SOP, FMEA, control plans, 8D and SPC, NPI and lean experience, and delivery of PFMEA and standard work instructions. Dahan Robotics' campus quality engineer profile lists 8D, 5Why, SPC, FMEA, APQP and PPAP. These are the standard skill sets of automotive and consumer-electronics quality engineers. Entry sits at vocational-to-bachelor level, pricing runs on systems fluency and a yield record, and the mobility path runs between car plants, EMS and robot factories.

Resumes and interviews differ across the fork. R&D candidates evidence themselves with papers, competitions, open-source work and iteration logs; manufacturing candidates with yield data, NPI projects led, audit records and defect-closure cases. Postings also write requirements such as 3D CAD proficiency and independent ownership of flash-and-test flows, conditions that almost never appear on an R&D resume. Screening both lines with one template misses two kinds of suitable people at once.

Put the two lines back into one cohort and an easy-to-miss number appears: the spread between job families is wider than the spread between degree tiers. Campus manufacturing is bachelor-heavy and campus R&D master-heavy, yet a process engineer two years out is back inside the entry band of R&D campus pay. Experience and systems capability now price almost as efficiently as the degree.

The fourth dimension is the city, and it needs a scope warning of its own: no whole-machine maker publishes city-tiered pay across two or more sites, so a city pay comparison cannot be constructed. Table 2 gives pay anchors only where a company announcement exists and leaves the rest blank with reasons.

Table 2: industrial concentration and pay anchors by region. Beijing and Liuzhou have no city-tiered published pay and are left blank, taking part through concentration indicators only; their prices can be read from Table 1 by job family and experience band. Data window: 2025 to October 2026. Sources: Nanfang Daily, Beijing E-Town, Guangxi industry authorities, campus announcements and company results.

Shenzhen's concentration measures the width of options. With 91,000 related firms and a half-hour industry circle, a robotics engineer there can move between machines, components, sensing and integration; the Nanfang Daily formulation is that upstream and downstream are one elevator ride apart. For candidates, option width beats a single-point pay premium; for employers, hiring inside a cluster is measurably more efficient.

Manufacturing function migrates into supply-chain radius. Liuzhou takes machine production on an auto-city supplier base and an ASEAN-facing position; Dongguan, Changzhou and Kunshan take quality and process roles inside the electronics and precision-manufacturing radius. The R&D end concentrates in university-dense cities, the manufacturing end where cost and logistics are friendly: one industry, two geographic curves.

The missing city pay matrix is itself information. Firms that put plants and headquarters in different cities without publishing city bands are saying that manufacturing pricing is set by job family and experience, with the city premium negligible for now. That matches a market still hiring centrally from headquarters.

Stack takt, pay bands, degrees and cities and five directional judgements fall out. Each is observable in public data on the basis used here, and each can be tracked forward on the same basis.

Trend one, the hiring structure moves from an R&D single centre to two parallel tracks. UBTech's R&D share is still 43%, but since Liuzhou started production, process, quality, equipment and supply chain roles are being configured against ten-thousand-unit capacity. The wafer-fab and car-plant precedent says the manufacturing track of a capital-heavy maker eventually outnumbers R&D; embodied intelligence is approaching that inflection, and counting the tracks as one overstates R&D supply and understates the manufacturing gap.

Trend two, automotive quality systems people become the contested cross-industry hire. Qiyuan imports IATF16949 outright, and postings write FMEA, PPAP and SPC into hard requirements. That capability stock sits in car plants, Tier-1 suppliers and EMS plants; robotics has no internal supply of it. Over the next two to three years most of the incremental supply of robotics process and quality people arrives by cross-industry movement, not from robotics degrees.

Trend three, pay tools move from cash only to cash plus equity. UBTech granted equity to 583 R&D staff against a cumulative loss background and opened a chief scientist search at RMB 15 million a year and up; head private companies campaign on listing expectations and option pools. Production-era competition runs on both layers at once, and comparing advertised salary without the equity layer systematically understates what head companies actually offer.

Trend four, the training side adjusts to production demand. The MIIT guidance calls for interdisciplinary, engineering-oriented and application-oriented talent. Employers move faster: Unitree opens 200 campus roles in one intake, and head plants build training floors inside production lines. The speed at which training cycles lock onto production cycles is becoming the real bottleneck of supply.

The bottleneck is structurally similar to the semiconductor one: curricula target R&D destinations while a new line needs people who can run assembly, calibration and testing from day one. Training floors inside plants and production stations in syllabi have begun to appear, and the channel is open, but scale needs two to three training cycles.

Trend five, falling machine prices and rising manufacturing pay run in parallel. The projected 49% price decline rests on supply chain localisation and line efficiency; unit cost falls while manufacturing engineer pay rises. The lithium battery and photovoltaic precedents both show process and quality pay outgrowing the industry average through the ramp, because the scarce input is the person who can lift yield. The next split in robotics pay happens inside the manufacturing track: line operators at market rate, the people who hold yield at a premium.

How a takt time becomes a hiring requirement: one prototype corresponds to a project organisation, a few dozen engineers around a single machine. Ten thousand units correspond to a factory organisation, and every term behind it, takt, yield, batch consistency, supplier audits, traceability, maps to a family of roles. Every second compressed from the 10-minute takt down to 2.5 was paid for in process and quality systems. The implication is direct: read a maker's takt and capacity under construction before its hiring plan, because a doubling of takt means the manufacturing profiles have to be rewritten.

III. Limitations, Tracking Indicators and What This Means for Employers

The judgements in this report are constrained by five limitations and should be cited together with them.

First, shipment accounting is fragmented: IDC, Frost & Sullivan and a third house put first-half 2026 global shipments at about 25,000, 36,000 and 19,100 units respectively, and this report takes only the directional conclusion without mixing values. Second, the pay anchors are few and cross-company: the manufacturing anchors rest on two campus announcements and two live postings, so cross-company comparison is directional only. Third, no city pay comparison can be constructed, and Table 2 leaves two cells blank for that reason.

Fourth, no industry shipment forecast is used as a denominator anywhere in this report; the 2030 projections in circulation come from consultancies with divergent scopes. Fifth, the per-employee payroll cost is a static all-staff figure: RMB 358,700 includes amortised equity cost and management and support staff, and converts to no specific role's take-home pay.

Four indicators can be tracked from public data alone, quarterly and on the same basis: the share of process, quality and equipment roles in head makers' campus announcements; actual output against planned capacity at plants already running, Liuzhou included; the ratio of the quality engineer pay centre to the technical engineer pay centre on campus, the most direct reading of whether manufacturing pricing is converging on R&D; and the relative growth of production headcount against R&D headcount in listed-company annual reports.

Table 3: four trackable indicators of production-era hiring structure. All four run on public data alone, recorded quarterly on the same basis; computations are this report's own and include no institutional forecasts.

For machine and component makers, write manufacturing profiles around systems capability. Hard requirements for process and quality roles should sit on FMEA, SPC, PPAP, IATF16949 and NPI record, with the degree gate relaxed to bachelor and the weight moved to project evidence; the search radius should span car plants, Tier-1 suppliers and EMS plants, because these candidates already speak the same quality language. Using campus pricing to chase systems capability returns a low hit rate; long-term incentives should extend from the research backbone to the manufacturing backbone, because once a line runs, the people who hold the yield are the most expensive to lose. Component-side employers should note that integrated actuator modules are moving component hiring from discrete-part assembly toward module integration and calibration, and profiles should be set at module level.

For foreign manufacturers in China, three readings deserve their own budget lines. The first is that the counterpart has changed. China humanoid robot companies price talent against equity-heavy structures and scale, and the industrial automation talent a multinational recruits is courted, at the same moment, by local champions holding listing expectations and option pools; a pure cash offer competes weakly, and role scope, project scale and equipment budgets become the workable levers.

The second is that IATF16949 fluency is portable evidence. PFMEA, control plans, PPAP files and audit records demonstrate systems capability independently of industry label, so assessing cross-industry hires means reading artefacts rather than checking sector boxes, and it cuts both ways: the same evidence base lets a car-plant quality engineer move into a robot factory at a premium. An IATF16949 quality engineer with a yield record is, at this phase, a harder hire than a robotics PhD.

The third is that supplier quality and localisation engineering headcount rise with the supply chain's global weight. As Chinese robot supply chains deepen and prices fall, foreign manufacturers auditing China-based supply need supplier quality engineers and localisation support staffed on the supply radius itself, in the Pearl and Yangtze delta belts. Humanoid robot manufacturing jobs in China concentrate along exactly those belts, and searches sited outside them run long: time to hire manufacturing roles of this profile runs on months, not weeks.

One calendar note closes the list. Head makers lock 2027 cohort offers in the autumn 2026 campus cycle, and the biggest engineering intakes are signed before international teams recruiting on a spring rolling basis have opened requisitions; any foreign employer competing for the same graduates needs its China hiring calendar synchronised to the local campus cycle, not to headquarters'.

As a robotics recruiter in China covering robotics, automotive and precision manufacturing, SunTzu China advises clients to bring manufacturing profiles and search maps into the plan at project approval, not after line installation; the value of a search partner at this phase lies in identifying systems capability, not resume volume.

Nature of the data: this report is an observational compilation rather than a survey, all data comes from public channels, and conclusions cover the sample only. Data window: September 2025 to 14 October 2026, with some policy and annual-report bases extending back to 2023. Pay basis: campus and posting figures are monthly and exclude equity, bonus and allowances; the per-employee payroll cost is total payroll divided by year-end headcount on a static basis; band centres are computed from published ranges and are not percentile statistics.

Job classification: five families, whole-machine R&D, process and NPI, quality, production, and control rows. Shipment basis: IDC, Frost & Sullivan and a third house are presented side by side and never mixed, and no industry forecast is used as a denominator of any ratio. City data: concentration indicators come from local government disclosures, customs statistics and company announcements; pay anchors appear only where a company announcement exists.

Appendix: main sources and metric definitions.

1. Ministry of Industry and Information Technology, humanoid robot development guidance (2023) and official statistics; government documents. (A)

2. UBTech 2025 annual results, Liuzhou plant disclosures and official postings; listed-company disclosures. (A)

3. IDC shipment statistics; Frost & Sullivan 2026 global humanoid robot market report; Omdia embodied robotics market radar; market research reports. (A/B)

4. Unitree and Dahan Robotics 2027 campus announcements, via university career channels; company announcements. (A)

5. Nanfang Daily, Beijing E-Town, Guangxi industry authorities, Shenzhen Customs, and live postings on recruitment platforms. (A/B)

Metric definitions: takt time is the interval between consecutive units leaving the line; band centre is computed from the two ends of a published range and is not a percentile statistic; average payroll cost per employee is total payroll cost divided by year-end full-time headcount.

Disclaimer: this report is compiled from public information and constitutes neither investment advice nor any commitment of employment terms; pay data are source-reported or platform-listed values, and actual terms are those agreed between employer and candidate.

SunTzu China

This report is part of a series on humanoid robot production talent. Clients seeking role profiles, search maps and pay benchmarking across whole machines, actuator modules and core components are welcome to reach the advisory team through the website; the series will update takt and manufacturing pay readings quarterly.