Start with the matcher, then use the sourcing report to decide which China rotary-joint architecture deserves supplier review, sample testing, or rejection.
Match a humanoid joint location, torque range, and backlash target to a China-sourcing shortlist. The output is an RFQ filter, not a substitute for bench validation.
Select all three inputs, then run the matcher. For unknown torque, choose the higher range and treat the result as a supplier-review path.
The China high-precision rotary-joint opportunity is strongest when the tool layer and RFQ discipline work together: use the matcher to narrow the architecture, then verify every precision, shock, thermal, and procurement claim.
Official policy and a dense robotics supply chain support faster supplier discovery, but a humanoid rotary joint still needs module-level validation before it becomes a production part.
Harmonic joints fit compact precision axes. Cycloidal, RV-style, and QDD paths deserve priority for leg, waist, and other load-bearing axes where shock and heat dominate.
A useful quote separates reducer, motor, encoder, driver, brake, bearings, housing, thermal path, and test method. Reducer-only backlash is not enough for humanoid procurement.
If a supplier cannot provide backlash-under-load, torque-speed, thermal, and life-cycle data, keep the vendor in prototype evaluation and avoid scaling the part across many joints.
| Architecture | Best For | Watch For |
|---|---|---|
| Harmonic / strain-wave | Arm, wrist, neck, sensor-pointing, compact precision axes | Reducer-level backlash claims, impact fatigue, heat, and limited backdrivability Validate: Measure complete-joint lost motion, repeatability, preload, thermal drift, and noise. |
| Cycloidal / RV-style | Hip, knee, ankle, waist, and other high-stiffness load paths | Weight, packaging size, measurable backlash, vibration, and supplier-specific geometry Validate: Test shock case, torsional stiffness, backlash repeatability, bearing load, and lubrication life. |
| Planetary / QDD | Early prototypes, backdrivable axes, hands, and cost-sensitive modules | Stiffness limits, audible gear noise, backlash growth, and holding torque Validate: Run duty-cycle wear tests and confirm controller compensation can handle compliance. |
| Integrated joint module | ODM prototype loops and teams that need motor, reducer, driver, and encoder in one package | Opaque component choices, thermal path, firmware limits, connector quality, and serviceability Validate: Request module-level datasheet, teardown detail, firmware interface, and replacement-part policy. |
This report avoids unsupported market-share and cost-reduction numerics. The evidence below shows why each sourcing criterion is reasonable and where public information stops.
| Claim | Evidence Use | Source | Type / Date |
|---|---|---|---|
| China humanoid robot policy timing | MIIT guidance published in November 2023 sets 2025 and 2027 milestones for humanoid robot innovation and industrial ecosystem development. Use this as a supply-chain timing signal, not as proof that any one supplier is qualified. | MIIT humanoid robot innovation guidance | Official policyPublished November 2023; checked July 26, 2026 |
| Performance-test vocabulary | ISO 9283 separates industrial robot performance criteria and test methods. For humanoid rotary joints, use the same discipline by recording measurement conditions for backlash, repeatability, path error, and thermal state. | ISO 9283 standard page | International standardStandard published 1998; checked July 26, 2026 |
| Harmonic-drive positioning role | Harmonic-drive and strain-wave suppliers position the architecture around compact precision and low or zero backlash. This supports shortlisting for arms, wrists, and necks, but module-level tests are still required. | Harmonic Drive technology reference | Manufacturer technical referenceChecked July 26, 2026 |
| Chinese harmonic reducer supplier signal | Leaderdrive publishes harmonic reducer and actuator product families for robot applications. Use public product pages to start the RFQ, then request exact backlash, torque, and life-test conditions for the complete joint. | Leaderdrive product catalog | Manufacturer catalogChecked July 26, 2026 |
| RV/cycloidal role in robot joints | Nabtesco positions RV precision reduction gears for robot joints that need compact packaging, rigidity, overload resistance, low backlash, and accurate positioning. This supports the cycloidal/RV-style comparison for load-bearing humanoid joints. | Nabtesco Corporation precision reduction gears | Manufacturer technical referenceChecked July 26, 2026 |
| Evidence limitation | Public sources do not provide a neutral 2026 benchmark that compares complete Chinese humanoid rotary-joint modules under identical torque, thermal, duty-cycle, and shock boundaries. Vendor data should therefore be treated as RFQ input until tested. | Page synthesis and sourcing guardrail | Evidence gapReviewed July 26, 2026 |
Limit: no public supplier-neutral benchmark found for complete Chinese humanoid rotary-joint modules tested under the same torque, thermal, impact, and duty-cycle boundary as of July 26, 2026. Use vendor claims to build the RFQ, then verify with samples.
A hybrid page should help the buyer act. These risks convert the report into RFQ questions, sample tests, and rejection criteria.
A supplier may quote reducer backlash while the robot needs complete-joint lost motion after assembly, preload, encoder mounting, and thermal soak.
Mitigation: Require module-level backlash or lost-motion curves with test torque, temperature, and preload stated.
Humanoid joints operate in compact envelopes where continuous torque can be far below peak torque once the motor, driver, housing, and cooling path are assembled.
Mitigation: Ask for torque-speed curves and temperature rise at continuous duty cycle, then test one sample on a dyno.
Load-bearing axes can see impacts that are not represented by static torque ratings or precision claims.
Mitigation: Define the landing or overload case in the RFQ and keep high-risk suppliers at prototype stage until impact cycling is measured.
A low module quote may exclude driver, brake, encoder, cable set, import cost, replacement parts, test fixtures, or supplier engineering time.
Mitigation: Compare landed cost per validated joint, not catalog price per reducer.
A joint that works for a bench demo may fail in a service robot, industrial payload case, outdoor environment, or repeated fall-recovery test.
Mitigation: Segment RFQs by joint location and duty cycle instead of scaling one actuator family across the whole robot.
Use the matcher result as the architecture hypothesis. The RFQ package below decides whether a China rotary-joint supplier can support humanoid validation.
| RFQ Field | Required Detail |
|---|---|
| Joint map | Location, lever arm, envelope, mass target, voltage, cable exit, and mounting interface. |
| Load case | Peak torque, continuous torque, target speed, duty cycle, fall or overload assumption, and holding/brake need. |
| Precision target | Backlash or lost-motion target, repeatability target, encoder resolution, and test conditions. |
| Thermal and life data | Torque-speed curve, thermal derating, temperature sensor access, lubrication, fatigue cycle assumption, and service interval. |
| Procurement boundary | Sample quantity, pilot volume, annual volume target, warranty, replacement-part plan, and export/import constraints. |
These supporting pages keep the sourcing workflow connected: architecture selection, actuator module scope, thermal review, RFQ packaging, and backlash test definition.
China can be attractive when a humanoid program needs supplier density, ODM iteration, integrated motor-reducer-driver modules, and sample speed. It is not an automatic quality signal. Shortlist suppliers only after comparing torque-speed curves, backlash test method, thermal derating, encoder placement, and lifecycle assumptions.
Leg joints usually need shock tolerance, stiffness, and thermal margin before minimum backlash. Cycloidal, RV-style, or QDD architectures are often better RFQ starting points than a generic zero-backlash harmonic drive. Final selection depends on robot mass, lever arm, landing case, and controller strategy.
Use harmonic or strain-wave joints for compact axes where low lost motion and smooth positioning matter, such as arms, wrists, necks, or sensor-pointing joints. Validate complete-joint backlash, not only reducer-level backlash, because bearings, housings, encoders, and assembly preload affect the final axis.
Some suppliers publish very low backlash or lost-motion specifications, but claims are comparable only when the torque, preload, temperature, measurement method, and whether the value is reducer-level or module-level are visible. Treat missing test conditions as an RFQ risk.
Request peak torque, continuous torque, torque-speed curve, thermal derating, backlash or lost-motion curve, torsional stiffness, reflected inertia, encoder resolution and placement, brake data, bearing load, IP rating if needed, duty cycle, and lifecycle test conditions.
ISO 9283 is written for industrial robot performance tests, not humanoid joint procurement. It is still useful as a vocabulary anchor because it separates accuracy, repeatability, path behavior, and test conditions. A humanoid RFQ should adapt that discipline at joint and subsystem level.
The common mismatch is comparing a reducer specification from one vendor with a complete actuator module from another. A complete humanoid joint also includes motor, encoder, driver, brake, bearings, housing, cable routing, seals, and thermal path.
Use the matcher conservatively by selecting the higher torque range, then ask the supplier to classify the joint after reviewing your lever arm, duty cycle, speed, robot mass, and fall case. Unknown torque should delay volume procurement, not necessarily sample exploration.
Integrated modules can reduce assembly risk and speed prototype loops, but they can also hide component-level limits. Ask for module-level tests and teardown-level details when the axis is safety-critical, load-bearing, or thermally constrained.
Validate one sample joint on a dyno before multi-joint procurement. Measure backlash before and after thermal soak, continuous-torque temperature rise, speed under load, controller stability, impact or overload behavior, noise, and connector durability.
Avoid paying for ultra-low backlash when the application mainly needs transparency, recoverable impact, or cost-efficient learning cycles. A lower-ratio QDD or planetary prototype can be the faster path if control compensation is acceptable.
Send a joint map with location, envelope, voltage, peak and continuous torque, target speed, duty cycle, cooling path, backlash target, encoder requirement, cable exit, brake need, expected impact case, validation standard, sample quantity, and target annual volume.
Move from architecture matching to supplier review. Share your torque, speed, duty-cycle, and validation requirements for custom humanoid rotary-joint sourcing.
