Short answer: Source robot reducers — harmonic (strain-wave) units and RV (cycloidal) units — by verifying consistency before negotiating price. The specification numbers are easy to print; what you are buying is a supplier's ability to hold backlash, stiffness and noise at the same level across hundreds of units and several years. That is tested with sample measurements, a pilot batch and factory questions, in that order. The principles and trade-offs behind the two families are covered in our guides to harmonic reducers explained and RV reducers vs harmonic drives; this article is about the buying.
Know the supplier types before you compare quotes
Quotes arrive from four different animals. The original manufacturers — the established Japanese, German and increasingly Chinese makers catalogued in the directory's core components section — sell engineering support and traceability at the highest prices. Their authorised distributors add local stock and first-line support, sometimes at little premium. Regional manufacturers offer the same interfaces at lower prices with more variable documentation. White-label and surplus channels offer price and risk in equal measure. Comparing a quote without knowing which animal sent it compares nothing; the first sourcing task is classification, and the second is asking each type the questions it would rather not answer.
Sample testing: measure the promises that matter
Order samples before volumes, and measure the properties your application actually feels. Lost motion and backlash, measured at the output under reversing torque, decide positioning quality. Torsional stiffness — how much the unit winds up under load — decides how your control loop behaves. Starting torque and no-load running torque reveal assembly and bearing quality. Noise and vibration across the speed range, and temperature rise under your duty cycle, predict the complaints you would otherwise meet in the field. Test more than one sample: a single golden unit proves the supplier can build one reducer, and your product needs them to build ten thousand alike. Record serial numbers and keep the reports; they become the reference for every later dispute about drift.
Lead times: what actually drives them
Reducer lead times stretch and shrink with the industry's humanoid and automation cycles, because the same capacity serves everyone. The honest drivers are casting and gear-cutting capacity, heat treatment, and the maker's order book — not the salesperson's calendar. Ask what the quoted lead time assumes, which stage of production your order joins, and what the supplier's delivery record looked like in the last busy period rather than in this quiet one. For production programmes, the standard protections apply with extra force: qualify a second source for the same interface early, hold buffer stock against the longest credible delay, and treat an implausibly short lead time from a new supplier as a quality question, not a bargain.
From samples to production: the ramp is a process
Between the sample that passed and the production that ships sits the pilot batch — a small production-quantity order, measured with the same tests as the samples, at the production price and through the production line. Discrepancies between sample and pilot performance are the most valuable information in the whole sourcing exercise, and they are only available before you are committed. Agree in this phase how changes are controlled: if the supplier alters a material, a bearing source or a process, how are you notified, and what re-qualification follows? Reducers fail programmes quietly, through drift nobody contracted against.
Documentation and traceability to insist on
Each shipment should arrive with what your own customers will one day demand of you: certificates of conformance against the ordered specification, the measured key characteristics per batch (or per unit, for precision grades), and serial traceability to production records. Ask how warranty claims are diagnosed — what evidence the supplier needs, where failed units go, and what turnaround is realistic to your country. These are the same acceptance habits applied elsewhere in the components market; our sourcing guide to ball screws and linear motion makes the parallel case for linear parts.
Contract points that outlive the price
Price per unit is the clause everyone reads; the programme is protected by the others. Specification freeze and change notification, as above. Capacity reservation or at least forecast-sharing mechanics for volume ramps. Warranty terms expressed in operating hours or cycles, with a defined remedy. And an end-of-life clause: precision reducers live in products that ship for a decade, so ask how long the exact model will remain orderable and what its designated successor is. A reducer is a small part of a robot's bill of materials and a large part of its reputation; source it like the second of those.
Frequently asked questions
What should I test on reducer samples before ordering in volume?
The properties your application feels: lost motion and backlash under reversing torque, torsional stiffness, starting and no-load running torque, noise and vibration across the speed range, and temperature rise under your duty cycle — on several samples, not one golden unit.
Why do reducer lead times vary so much?
Capacity in gear cutting and heat treatment is shared across the automation and humanoid industries, so lead times follow those order books. Ask what stage of production your order joins and what the supplier delivered during the last busy period, not this quiet one.
What is a pilot batch, and why does it matter?
A small order at production quantity conditions — production line, production price — measured with the same tests as the samples. Differences between sample and pilot performance are the cheapest warning you will ever get, and they only exist before you commit.
What contract terms matter besides price?
Specification freeze with change notification, forecast and capacity arrangements for ramps, warranty expressed in hours or cycles with a defined remedy, and an end-of-life clause stating how long the model stays orderable and what succeeds it.