> ## Content Index > Fetch the complete content index at: https://www.metrology.com/llms.txt > Use this file to discover other available public pages before exploring further. # From demo to factory - scaling robot production and the actuator bore gauging gap - URL: https://www.metrology.com/from-demo-to-factory-scaling-robot-production-and-the-actuator-bore-gauging-gap/ - Published: 2026-08-03T04:20:54.000Z - Updated: 2026-08-03T04:56:02.000Z - Description: A practical guide for engineering leads and investor due diligence reviews - Author: The Datum - Tags: The Datum ## The Demo-to-Factory Trap A robotic actuator is a joint module: a frameless torque motor, a precision reducer, an output bearing, two encoders and drive electronics, packed into one housing. Dimensionally it is a stack of concentric bores that have to agree with each other, since the motor sits in one, the reducer sits in another and the output bearing sits in a third. A humanoid carries 28 or more joints, so a $20,000 platform is running an $11,000 actuator budget at roughly $400 per joint, and actuators account for approximately 56% of the bill of materials. The tolerances on those bores are single-digit microns in places, because reduction ratios between 30:1 and 50:1 leave the designer almost no envelope to work in. Disagreement between the bores shows up as scrap, rework and unexplained runout on the largest line in the cost model. At twenty demo robots that is 560 actuators to get right, and at five hundred a month it is 14,000. > The punchline. Your ability to scale and meet customer and investor expectations means you need to get really good at measuring 4 concentric diameters to plus or minus a few microns. Most humanoid and legged programmes design the actuator and assemble it in-house, buying the reducer and motor as components, since catalogue joint modules arrive 30% to 50% heavier than the joint tolerates. First units get built on a bench, with bearings selected from a batch and shims cut until preload and runout come in. That is correct at twenty units, where formalising inspection against an unfrozen design wastes money on features that will change. It stops working at five hundred a month, because hand selection now needs 14,000 conforming parts and without an inspection gate the tail ships by default. Teams usually read the result as a firmware problem. Unit-to-unit variation in friction and torque ripple pushes control engineers into compensating in software, and the compensation only holds for the batch it was tuned on. A bore rejected at goods-inward costs the price of the part, while the same deviation found at end-of-line test costs the assembly plus teardown labour, a 10x to 100x multiple. Feedback time sets how many units are affected, since a CMM queue running hours to days runs while the machine keeps cutting. The punchline. Your ability to scale and meet customer and investor expectations means you need to get really good at measuring 4 concentric diameters to plus or minus a few microns. --- ## The supplier position makes things harder for start ups and scale ups Supply economics makes all of it harder. Harmonic Drive Systems held around 85% of the harmonic reducer market in 2023 and Nabtesco claims roughly 60% of medium-to-large joint reducers, and outside the vertically integrated Chinese programmes almost nobody makes their own strain-wave units. There is no practical second source, and 20 to 30 week lead times turn a rejected batch into a big issue. Rejecting a shipment carries real commercial cost, since the supplier has other customers and a small buyer has limited ability to escalate. Finding a problem at week 4 leaves the options to rework, resequence or requalify open, while finding the same problem at week 24 usually means a missed milestone. Further, as more "standard" assemblies appear the market moves toward integrated joint modules combining reducer, frameless motor, encoders and thermal sensors in one black box housing. That leaves the mounting register, the output flange pilot, the hollow shaft ID and your own housing bores. Most stack-up failures occur at that interface, and responsibility for them is regularly disputed between supplier and buyer. --- ## The Bores That Decide It The working list follows, with each feature described by the failure it produces and the gauge configuration that reads it. The first group covers the motor side, which is common to every joint module. The second covers the reducer, where the features depend on whether the design uses a strain-wave, planetary or RV unit. Three-point head sizes across our range run 2 mm to 40 mm, with anvil geometry configured to the feature rather than to the nearest standard size. ![](https://storage.ghost.io/c/97/2a/972a49a2-bf56-480d-bfc6-da726d4924be/content/images/2026/08/actuator-bore-gauging-callouts--1-.svg) ### 1) Motor side ### 1.1) Stator housing bore Robot joint motors run air gaps of 0.3 to 0.5 mm because torque density scales inversely with gap. A housing bore 20 µm out of round consumes 4% to 7% of the design gap on one side, producing unbalanced magnetic pull that appears as cogging torque, bearing side-load, acoustic noise and efficiency loss. Ovality does more damage than size error, since a bore uniformly 15 µm oversize shifts the gap while a bore 15 µm out of round modulates it once per revolution. **Gauge configuration.** Standard three-point head with extended reach, taking readings at both ends plus the midpoint to separate taper from ovality. In most designs this is machined in-house, so it is the first place to put a gauge at the machine. ### 1.2) Lamination stack internal diameter The shell bore and the stack bore are different features, and it is the stack ID that sets the air gap. A stack measured on the bench will read differently once it is pressed or bonded into the shell, because the laminations relax against the housing and take up its form. Measuring only the empty shell gives a number that is correct about the housing and wrong about the motor. **Gauge configuration.** Extended-reach head used after insertion rather than before, on 100% of units. This is one of the few features where the measurement has to happen at a specific point in the assembly sequence rather than wherever it is convenient. ### 1.3) Rotor press-fit ID and magnet ring bore The rotor ID that presses onto the shaft or the wave generator hub carries two failure modes at once. Ovality produces an unbalanced air gap on the running side and a dynamic imbalance that shows up as vibration at speed, and on bonded magnet rings the adhesive cure moves the geometry after the part was last measured. **Gauge configuration.** Standard head sized to the fit, measured before and after bonding. The second reading is the one that matters and the one most often skipped. ### 1.4) Encoder hub bore The encoder hub bore is usually 8 mm to 20 mm and its eccentricity converts directly into angular position error. A 10 µm offset on a 20 mm hub is arcminutes at the output, which the control loop cannot distinguish from real joint motion and will attempt to correct. **Gauge configuration.** Small-bore head with a dedicated setting ring. The tolerance is tight relative to the diameter, so the setting ring should not be shared with other features in the same size range. ### 1.5) Rear cover bearing seat and spigot register The rear cover carries the non-drive-end bearing seat plus the spigot that locates the cover to the shell, typically H6 and H7 respectively. Concentricity between the two sets rotor alignment at the rear, and error here loads the front bearing in a way that reads as a bearing quality problem. **Gauge configuration.** Blind-bore head, since the seat is usually closed at one end, measured on the same setting ring as the mating shell feature. ### 1.6) Hollow shaft internal diameter Cable routing through the joint means a deep through-bore, often at a length-to-diameter ratio above four. Two-point measurement is slow and unreliable in this geometry, because the operator has to rock the instrument searching for a maximum reading in a bore they cannot see into. **Gauge configuration.** Deep-bore head on an extension, with the display at the operator's end while the contacts work at depth. Three-point measurement matters most here, since the operator has no line of sight to the feature being measured. ### 2) Reducer side ### 2.1) Reducer mounting register and output flange pilot These are the interface features between the bought-in reducer and your housing, and they are the ones you can verify on receipt. Concentricity between the register and the output flange sets how much of the reducer's rated accuracy survives installation, and a 20 µm error here is indistinguishable in the field from a reducer that was out of specification when it shipped. **Gauge configuration.** Standard head matched to the register diameter, run against the same setting ring used on the mating housing bore so both halves of the fit sit on one datum. Measuring the two features on different instruments introduces a bias that is usually mistaken for an assembly fault. ### 2.2) Crossed roller bearing seat The output bearing seat is typically specified H6, which at Ø62 mm is a band of +0 to +19 µm. The seat provides both the interference fit and the geometric reference for the entire output flange, so ovality transfers almost directly into flange runout and is then multiplied by link length. On a 1.5 m arm, 30 µm of shoulder bore error becomes visible positional error at the wrist. **Gauge configuration.** Standard head sized to the seat, with shortened anvils where the seat is a narrow land rather than a full-length bore. On narrow seats a catalogue instrument will often contact the chamfer or relief groove and return a reading taken off the wrong surface. ### 2.3) Planet carrier bores Relevant where the joint uses a planetary first stage rather than a strain-wave unit throughout. A planet carrier holds each planet gear on a pin located in its own bore, and a 10 µm size spread across three of those bores routes a disproportionate share of torque through one planet. The fatigue failure that follows occurs at a load the design nominally survives, and in the field it is usually recorded as a design deficiency when the cause is dimensional. Bore-to-bore pitch matters equally and is a position measurement, so it stays on a CMM or a fixture while the gauge covers size and roundness. **Gauge configuration.** Blind-bore head, since carrier bores are frequently stepped or closed at one end. Blind-bore anvils measure to within a fraction of a millimetre of the bore floor, which is where the bearing sits and where the load is carried. ### 2.4) Cycloid disc and pin bores Relevant to RV and cycloidal designs rather than strain-wave. A mid-size RV unit carries 30 to 40 needle roller pin bores at 8 to 20 mm diameter, generally to IT5, a band of about 8 to 9 µm at those sizes. With the eccentric cam bores and the crankshaft bores through the carrier, one reducer presents more than forty critical internal diameters. Inspection volume per part is the practical constraint, since a feature count that high puts CMM inspection out of economic reach at rate. **Gauge configuration.** Small-bore heads in the 8 mm to 20 mm range, set up as a bench station with the setting ring alongside so an operator works through forty features without re-zeroing between each. Self-centring geometry supports that throughput, since there is no rocking or searching per feature. ### 2.5) Wave generator bearing seat The tightest feature in most actuators, frequently held to a total band of around 5 µm. The bearing is deliberately deformed in service, so the bore locating it has no margin left for manufacturing error and whatever geometry the bore contributes adds to the deformation the design already imposes. Loss of ratcheting margin presents as sudden torque loss under peak load rather than gradual degradation. **Gauge configuration.** Small-diameter head with carbide contacts and a dedicated setting ring at the tightest available resolution. This is measured during reducer manufacture rather than at goods-inward, since the feature is inaccessible once assembled. Buyers rely on supplier records for it, which is why the interface features above carry so much of the verification load. ### 2.6) O-ring grooves and internal recesses Sealed joints carry internal grooves whose diameter and width control seal compression, and a groove 30 µm out on diameter shifts squeeze enough to change seal life materially. These features are invisible to any instrument that cannot pass the entry diameter and then expand into the groove. **Gauge configuration.** Custom groove anvils designed to your groove specification. The parameters are the measured diameter, the entry diameter, the groove width, the depth to the groove, the clearance depth below it and the bottom clearance diameter, with the design check being that measured diameter minus entry diameter falls within head range. --- ## Why Two-Point Measurement Will Not Get You There A two-point instrument cannot detect odd-lobed form error. A shape with an odd lobe count holds constant diameter in every direction while being demonstrably not round, so the gauge reads it as conforming and the part passes inspection. Three-lobed form is the characteristic signature of three-jaw chucking and collet distortion, which covers the most common workholding methods in precision turning. A part clamped in a three-jaw chuck, bored, then released relaxes into a three-lobed shape whose diameter reads perfectly and whose bearing seat still produces runout. Honing tends to produce five-lobed and seven-lobed patterns for the same reason of periodic support, so the finishing operation does not remove the problem. Lobed form is the failure that carries straight through the transition to volume. At prototype scale a lobed bearing seat gets absorbed by selection and shimming; at volume it produces a population of joints with runout that the inspection records cannot account for, since those records show every bore in tolerance. Three contacts at 120° register the lobed condition rather than averaging through it. Self-centring geometry adds a second benefit, in that the instrument finds true diameter without rocking or searching. Against a 5 µm band, operator technique is frequently the largest single contributor to measurement uncertainty. --- ## Building the Regime: A 6 Step Practical Sequence The steps below run in order and are achievable inside a quarter. Total cost is less than a single scrapped batch of reducers. **Step one — write the critical bore list.** Take the joint module drawing and mark every internal diameter, then sort them by whether they gate function or only fit. Function-gating features get 100% inspection and everything else goes to sampling once capability is shown, and for most joint modules the first group runs to four or five features rather than forty. **Step two — select instruments against gauge R&R, not against diameter.** Convention holds that measurement system variation should consume no more than 10% of the tolerance band, with 10% to 30% acceptable under justification. Against a 5 µm band that means a system reproducible to 0.5 µm across operators and shifts, which most general-purpose instruments do not achieve in a working shop, so run the R&R study before you commit rather than after a customer rejection. **Step three — put the gauge where the metal is cut.** A CMM produces data on a schedule set by the queue in front of it, which in most shops runs from hours to days, while the machine keeps cutting. Gauging at the machine turns a process drift from a scrap event into a tool-offset adjustment. Where machining is outsourced, this splits in two: 100% inspection of critical features at goods-inward, plus an at-machine gauging requirement written into the contract shop's purchase order with the instrument and setting ring specified. A supplier measuring on your datum removes the bias that otherwise presents as an untraceable assembly fault. **Step four — set the sampling plan in writing and tie it to capability.** Start at 100% on the critical list and reduce only when Cpk supports it, with the rule written down rather than held by whoever runs the cell. Where the rule is not written down, sampling frequency tends to drift downward under schedule pressure with no record of when it changed. **Step five — establish calibration and recall.** Setting rings drift and instruments get dropped, so a gauge outside its verification interval carries no evidential weight whatever the display says. Quarterly verification against a traceable standard is a defensible default for actuator work, tightening to monthly where the gauge is in continuous production use, and the evidence chain matters as much as the calibration itself, since the first question in a supplier dispute is whether the buyer's measurement can be defended. **Step six — enforce thermal discipline.** Steel expands roughly 11.7 µm per metre per °C, so a Ø60 mm bore shifts about 0.7 µm per °C. A 7 °C difference between the gauge, the setting ring and the part consumes the entire tolerance band on the tightest features, which makes soaking parts and instruments together a hard requirement rather than good practice. --- ## Why the Tolerances Are This Tight The bands above are set by the design constraint rather than by convention. The governing metric for a humanoid joint is specific torque in newton-metres per kilogram, and it dominates because the actuator carries itself, unlike an industrial arm bolted to a floor where extra shoulder mass costs nothing beyond a larger base casting. Mass added at the elbow must be carried by the shoulder, then the torso, then the hip and the knee, so a 500-gram penalty at the wrist propagates into several kilograms of structure and battery by the time it reaches the ankle. Tesla, Figure, Apptronik, Agility, Unitree and Boston Dynamics all design their own actuators rather than buying modules for this reason. Chasing specific torque forces high reduction ratios, rising to 160:1 where static holding torque dominates. Each step up shrinks the envelope the designer has to work in, and the bore tolerances compress with it. High ratios also make the joint mechanically opaque, so the motor cannot sense external forces through the friction and inertia of the gear train. The design adds output-side torque sensing plus a software layer to synthesise the compliance direct-drive joints get for free, and once the joint depends on strain-gauge output rather than motor current, mechanical variation in the bore locating that element becomes control-loop error. --- ## The Supply Side, and Why It Gets Harder Before It Gets Easier McKinsey's humanoid supply-chain analysis ((as an example) identifies harmonic drives as the clearest constraint in the actuator stack, on the basis that production is precision-bound and capital-intensive, requiring dedicated tooling, long qualification cycles and the metrology infrastructure to support both. Reducer capacity cannot be commissioned in months the way electronics capacity can. Demand arithmetic explains the current capacity build-out. A single humanoid takes 20 to 40 harmonic reducers against six to twelve actuators for a conventional industrial arm, so one humanoid consumes the reducer content of three to five industrial robots. The International Federation of Robotics recorded more than 590,000 industrial robot installations globally in 2024, up 12.4%, while Omdia has humanoid shipments rising almost 480% in 2025 to 13,318 units with a projection of 2.6 million by 2035. > A buyer with a documented inspection regime and defensible measurement records is easier to deal with than one rejecting batches on a functional test failure it cannot localise. That projection implies between 52 million and 104 million precision reducers a year against current global capability measured in low single-digit millions. Capacity is being added, with Zhejiang Shuanghuan bringing on a Suzhou plant in 2026 described at 500,000 units annually and Leaderdrive growing 2025 revenue 47% to 570.7 million yuan, but the gap will not close within the timescale of a near-term production ramp. For a startup scaling now, lead times are likely to lengthen before they shorten, and allocation depends partly on how a supplier rates you as a customer. A buyer with a documented inspection regime and defensible measurement records is easier to deal with than one rejecting batches on a functional test failure it cannot localise. --- ## What It Costs The full regime for a joint module programme is a small number of gauges, a set of setting rings, a written plan and a calibration interval. Set against a single scrapped batch of reducers on a 20 to 30 week lead time, or a quarter of control engineering spent compensating for mechanical variation, the payback period is short. Teams that get through the ramp generally treat measurement as production infrastructure rather than as a quality department function. Write the critical bore list before you need it, buy instruments capable against the tightest band rather than the most common diameter, and put them where the parts are made. --- ## About metrology.com metrology.com designs and manufactures custom three-point internal bore gauges with standard heads from 2 mm to 40 mm and bespoke anvil geometry for features catalogue instruments cannot reach. The design office works from customer drawings, covering blind bores, recessed seats, O-ring grooves, deep hollow shafts and non-standard sizes, and returns a gauge specified against the actual feature. If you are scaling actuator production and have not yet written down which bores are critical, start with the list above. Send us the joint module drawing and the design office will come back within 24 hours with the features that warrant measurement at the machine and the head configuration that reaches them.