We sell integrated joints, so read this knowing where our bread is buttered. But we've also sat on the other side of this decision (it's how we ended up building them) and the honest answer is that both paths are right for different teams. What follows is the cost structure nobody puts on a quote.
What "build your own" actually involves
The parts list looks simple: component strain wave gear set, frameless torque motor, motor-side encoder, output encoder if you're wise, servo drive, brake if the axis holds against gravity. The engineering list is where the year goes:
- Alignment and preload. A strain wave gear set is not assembled so much as installed to tolerance: wave-generator concentricity, flexspline mounting flatness, and bearing preload all land directly on backlash, life, and noise. Get it slightly wrong and you'll meet the failure modes early, with no spec sheet to blame, because you are now the manufacturer.
- The thermal path. A frameless motor's rating assumes a heat sink: your housing. Designing the conduction path, then derating honestly for it, is a real thermal-engineering task.
- Feedback and commutation. Mounting and calibrating two encoders, aligning commutation, tuning current/velocity/position loops for a compliant transmission: weeks, per axis design.
- Verification. When it's built: how do you know the backlash, repeatability, and rated torque you designed for are the ones you got? A test bench is its own project.
What the integrated joint buys, and what it costs
An integrated joint collapses that list into a flange and a connector: reducer, motor, dual encoders, drive electronics, and optional brake in one verified housing. A Ø40 mm unit at 186 g carries all of it, with the backlash measured on the bench and logged to its serial number before it ships. The engineering you buy is the boring, load-bearing kind: someone else's alignment fixtures, thermal testing, loop tuning, and QA history.
The costs are real too: unit price is higher than the summed components; you accept the maker's envelope, connector, and firmware behaviour; and deep customisation (an odd hollow-shaft diameter, an exotic winding) is a negotiation, not a CAD edit.
When building your own is the right call
Three cases where we'd tell you to roll your own, without hesitation:
- The envelope is the product. If the joint must be a shape no catalogue offers — flat pancake inside a wheel hub, extreme hollow bore, integrated into a structural casting — component sets exist precisely for you.
- Volume with a captive team. At thousands of units a year with dedicated actuator engineers, the integration cost amortises and the per-unit saving compounds. This is why high-volume humanoid programs build in-house.
- The actuator is your differentiation. If your product's moat is joint performance itself (research platforms, prosthetics, aerospace mechanisms), outsourcing it outsources the moat.
When integration wins
- Machine builders whose value lives above the joint: the arm, the process, the software. Every month spent on actuator integration is a month not spent on the thing customers pay for.
- Low and medium volumes, where six engineer-months never amortises.
- Schedules with a deadline. A catalogue joint is a lead time; a custom joint is a development program with a lead time at the end of it.
- Anything needing a paper trail. Serialised test records per unit beat "we tested the prototype" in every audit.
The decision, compressed
Count your annual volume, your spare actuator-engineering capacity, and the distance between your envelope and a catalogue frame. Two of three favouring custom? Build. Otherwise buy the joint and spend your engineers where your customers can tell the difference. And if the sticking point is a spec — a bore, a winding, a bus — ask before you conclude; the space between "catalogue" and "custom" is where a lot of quotes actually land.
Related: Second-sourcing a reducer · One encoder or two?.