Humanoid Robot Joint Motors: Which Lamination Choices Improve Torque Control

Key Takeaways

• Bonded lamination stacks reduce measured core loss by 8–12% versus welded stacks and suppress 6th/12th harmonic cogging torque components — a pass/fail difference when the spec reads < 0.5% cogging.
• Different joints need different steel: 0.20–0.27 mm thin-gauge NOES per EN 10303 (≤ 15 W/kg @ 1.0 T/400 Hz) for shoulder and hip; 0.10–0.15 mm ultra-thin NOES or medium-Ni alloy for wrist and finger actuators.
• Tooth-tip chamfer consistency of ±0.03 mm across all stator teeth is more impactful on cogging than the chamfer dimension itself.
• Thin-gauge NOES at 0.20 mm handles the broadest range of humanoid joint motors; Co-Fe rarely justifies its 10–20× cost premium outside bipedal leg joints.

A humanoid robot’s shoulder has to hold up an arm while the wrist threads a needle. Same machine, wildly different motor problems. And at the bottom of both — literally, physically — sits a stack of stamped steel sheets thinner than a business card.

We build those stacks. The frameless PMSM and high-pole BLDC torque motors inside humanoid joint actuators are among the most demanding stator lamination stack applications we’ve shipped. Here’s what the production data actually shows about which lamination choices improve torque ripple, cogging torque, and smooth motion control — and which choices waste money.

Causes of Torque Ripple in Humanoid Joint Motors

Torque ripple in a joint motor shows up as jerky motion in the robot. The control loop can compensate for some of it, sure. But the electromagnetic source of that ripple — cogging torque, harmonic distortion, uneven flux distribution — gets baked in at the lamination level. Bad steel choice, sloppy tooth geometry, wrong stacking method, and the best FOC algorithm in the world won’t save you.

Joint motors for humanoid robots — the kind driving shoulder rotary actuators rated at 40–100 Nm peak, or knee joints pushing 100+ Nm — are almost always frameless PMSM or high-pole-count BLDC designs. No housing. No bearings of their own. The stator stack gets pressed directly into the robot’s structural joint housing. Any dimensional error in the stack becomes a concentricity error in the motor. Which becomes a torque ripple source.

The simulation doesn’t know your welds are shorting laminations together.

We’ve seen customers come in with beautiful FEA results, their model showing < 0.3% cogging at rated torque, and then lose 30–40% of that predicted performance because the stack wasn’t flat enough or the bonding method stressed the steel. The gap between simulation and reality, in joint motors, is almost always a lamination problem.

Lamination Material Selection by Joint Function

Not every joint in a humanoid robot needs the same lamination steel. Treating shoulder, elbow, and wrist motors identically is a common first-pass mistake that costs either money or performance — usually both. Current-generation humanoid platforms run 28–40+ actuators across the body, and the torque, speed, and precision requirements differ radically from joint to joint.

Shoulder and Hip Joints

These carry the heaviest loads. Continuous torque requirements run from 40 Nm up past 200 Nm depending on the robot’s mass. The motor operates at relatively low speeds but must sustain high current densities for extended periods, so thermal performance matters.

For these high-load, moderate-frequency joints, we typically recommend 0.25–0.35 mm non-oriented electrical steel with silicon content around 2.5–3.0%. At the thin end (0.25–0.27 mm), the applicable standard is EN 10303 / IEC 60404-8-8 — the thin-gauge, medium-frequency specification — with grades like NO25-13 (0.25 mm, ≤ 13 W/kg @ 1.0 T/400 Hz) or NO27-15 (0.27 mm, ≤ 15 W/kg @ 1.0 T/400 Hz). For the 0.35 mm option, you move into the EN 10106 / IEC 60404-8-4 standard where grades like M270-35A (0.35 mm, ≤ 2.70 W/kg @ 1.5 T/50 Hz) apply.

Why the thickness split matters: a 10-pole-pair motor at 300 RPM runs at only ~50 Hz fundamental. At that frequency, the eddy current penalty difference between 0.25 mm and 0.35 mm is modest. Going below 0.20 mm for shoulder joints is rarely justified — you’re paying for thin gauge without capturing proportional loss reduction at these low electrical frequencies.

The permeability needs to be high because the motor design will push flux density to 1.6–1.7 T in the teeth at peak torque. Above that and you start saturating, which distorts the back-EMF waveform and feeds torque ripple straight into the output.

Elbow and Knee Joints

Mid-range torque (10–80 Nm), higher dynamic requirements. These joints accelerate fast and change direction often. The lamination priority shifts from raw thermal endurance toward low hysteresis loss and high permeability at moderate induction levels (1.0–1.4 T operating range).

We’ve had good results with 0.20 mm grades — specifically NO20-12 per EN 10303 (≤ 12 W/kg @ 1.0 T/400 Hz). The key insight from our production data: when you get the permeability right at the actual operating flux range (not the peak), the back-EMF linearity improves measurably. That feeds directly into cleaner current control. The servo loop gets smoother torque to work with.

Wrist and Finger Actuators

Small stators. Fine teeth. Very tight slots. Torque requirements are modest (1–20 Nm) but the precision demand is extreme — these are the joints handling manipulation tasks, like the 22-DOF dexterous hands on current-generation platforms, where 0.1° position error matters.

Here we push into 0.10–0.15 mm ultra-thin NOES grades (NO10 or NO15 per EN 10303) or, for certain high-end programs, nickel-iron alloys in the 40–50% Ni family.

The Ni-Fe option gives phenomenal permeability ($\mur$ > 50,000 at low field) and almost zero cogging at the low flux densities these tiny motors operate at. The tradeoff: significantly higher material cost, lower $B{sat}$ (≈ 1.5 T for 48–50% Ni grades), and annealing requirements that depend on the specific alloy composition:

• 40–42% Ni grades (controlled-expansion types adapted for magnetic use): anneal at 850–1000°C in a protective atmosphere (N₂, dissociated ammonia, or dry H₂). These are the more forgiving option.
• High-permeability 49% Ni grades (optimized for maximum $\mu$ and minimum coercivity): require 1100–1200°C in dry hydrogen (dew point below −40°C), with controlled furnace cooling at ~60–100°C/hour. Skip this and permeability can drop 50–80% from the fully annealed state.

In a wrist motor that weighs 80 grams total, the cost premium on the steel is negligible relative to the robot’s overall BOM. The annealing, however, is not trivial — pick the grade that matches the magnetic performance you actually need, not the most exotic option available.

Worth noting: Ni-Fe laminations are sensitive to stamping stress. We prefer to laser-cut these and follow up with the appropriate anneal cycle. Progressive die stamping of Ni-Fe is possible — and for 0.15–0.20 mm thicknesses some programs do run it successfully — but the tooling has to be optimized for the material’s softness and ductility, and the post-stamp anneal becomes even more critical to recover properties lost to cold work.

Lamination Material Comparison for Humanoid Joint Motors

Core loss values represent maximum guaranteed values per EN 10303:2015 and EN 10106:2015 where applicable, or verified ranges from our incoming coil Epstein testing for non-standard grades.

To clarify the overlap at 0.20 mm: this gauge sits right at the boundary. For joints where the fundamental electrical frequency stays below ~100 Hz (most shoulder/hip applications), 0.25 mm or even 0.35 mm captures most of the loss reduction and is easier to stamp. For elbow/knee joints with higher dynamic requirements and frequencies reaching 200–400 Hz, 0.20 mm is the sweet spot. We default to 0.20 mm when a motor design team hasn’t yet locked the gauge, because it gives the widest margin across the range of operating conditions typical in humanoid joints.

Lamination Assembly: Why Bonding Beats Welding for Cogging Torque

The way you hold laminations together isn’t just a structural decision. It’s an electromagnetic one. This is where we see the most “free performance” left on the table by teams that get the steel right but botch the stacking.

Welding

Laser or TIG welds along the outer diameter of the stack create localized short circuits between adjacent laminations. The heat-affected zone degrades the insulation coating and increases interlaminar conductivity. In our testing on 0.20 mm NOES stacks (NO20-12 grade), a four-seam laser weld added roughly 8–12% to measured core loss compared to the same stack bonded with adhesive.

That extra loss isn’t uniformly distributed. It concentrates near the weld lines. Depending on weld placement relative to slot positions, this creates asymmetric heating and introduces additional harmonic content into the flux distribution. We’ve measured it on an FFT of the cogging waveform — the 6th and 12th harmonic components increase noticeably on welded stacks versus bonded ones.

For industrial motors, nobody cares. For a joint motor where the spec says cogging < 0.5% of rated torque, it can be the difference between pass and fail.

Interlocking

Better than welding from an electromagnetic standpoint — no heat damage. But the interlock dimples create local deformations in the steel. Each dimple is a point of increased residual stress, which raises local hysteresis loss.

In small-diameter stators (anything under ~60 mm OD, which covers most wrist and elbow joints), there’s often not enough back-iron to place interlocks without affecting the magnetic circuit. We’ve seen cases where interlock placement in the yoke region of a 40 mm stator caused measurable flux density asymmetry in the airgap.

Adhesive Bonding (Recommended)

Self-bonding varnish (backlack) — classified as C-3 type per EN 10342 / IEC 60404-1-1 / ASTM A976 insulation coating standards — or post-stamp adhesive application produces stacks with:

• No interlaminar short circuits
• No residual stress points
• No magnetic circuit disruption
• Natural frequency 75%+ higher than interlocked stacks of equal geometry
• Vibration damping that welded/interlocked stacks can’t match

Our adhesive layer thickness: 2–5 μm per interface. This keeps stacking factor above 97%. Thicker adhesive layers eat into active steel fraction and can reduce torque density by 2–3%.

Bonded stacks also run quieter. The adhesive between layers damps the high-frequency buzz that welded or interlocked stacks transmit into the robot’s structure. In a collaborative robot working near humans, audible hum from motor cores is a real UX problem. Bonding kills it.

Tooth Geometry and Slot/Pole Design: Where Stamping Precision Gets Tested

Fractional-slot concentrated winding designs with high pole counts (16+ poles) are standard for joint torque motors. Common pairings like 12-slot/10-pole or 24-slot/22-pole inherently suppress low-order cogging harmonics. But the lamination still has to execute the geometry to spec.

A few things we’ve found matter more than the textbook suggests:

Tooth tip chamfering. We routinely add 0.2–0.4 mm chamfers to the tooth tips on stator laminations for robotics motors. In FEA, this reduces cogging by 15–25% on typical high-pole designs. But the improvement only materializes if the chamfer is consistent across all teeth to within ±0.03 mm. Inconsistent chamfers can actually increase cogging because they introduce geometric asymmetry that the fractional-slot design wasn’t meant to handle.

Burr height control. On 0.20 mm laminations, our production spec is ≤ 10 μm burr height. Every micron of burr is a potential interlaminar contact point that degrades insulation and creates eddy current paths. On thinner material (0.10–0.15 mm), the burr-to-thickness ratio gets aggressive fast.

For gauges below 0.15 mm, we typically switch to fiber laser cutting — primarily because it gives us tighter burr control and eliminates die wear as a variable. Progressive die stamping of 0.10 mm NOES is feasible (the material can handle slit, shear, and punch operations at room temperature), but maintaining consistent burr height below 10 μm across a full production run requires aggressive die maintenance schedules that most programs find impractical. Your mileage will vary depending on volumes and geometry complexity.

Slot opening width. Narrow openings reduce cogging but make winding harder and trap heat. The sweet spot for joint motors in the 40–80 mm stator OD range: typically 1.5–2.5 mm. The lamination tooling has to hold this dimension to ±0.02 mm across every slot, or the harmonic suppression from fractional-slot design gets partially negated.

Skewing: When and How Much

Skewed lamination stacks reduce cogging torque by spreading the magnetic interaction over a larger angular range. It works. But it’s not free.

A continuous skew of one slot pitch virtually eliminates the fundamental cogging component. It also reduces average torque by 1–3% and complicates winding.

For high-pole joint motors with fractional-slot designs, most programs request half-slot or partial skew — enough to knock down residual cogging without sacrificing meaningful torque. We implement this as stepped skew using 2–4 sub-stacks rotated relative to each other.

Our standard capability: 2-step or 3-step skew with angular accuracy of ±0.3° per step. For tighter specs: 4-step configurations at ±0.15° — requires custom fixturing and adds cost.

One interaction that doesn’t get discussed enough: skew and stacking method. Welded stacks with stepped skew develop stress concentrations at each weld-step interface. Bonded stacks handle the slight angular offset cleanly because the adhesive accommodates it without hard contact points.

Amorphous and Nanocrystalline Materials: Why Not (Yet)

We get asked about amorphous metal ribbon for joint motors occasionally. The core loss numbers are spectacular — at ~0.025 mm thickness, loss drops by 70–90% versus NOES at comparable conditions. The practical problems are significant: – Stacking factor drops to 80–85%. You lose active magnetic material because the ribbon is ~25 μm thick with proportionally more coating/air per unit height. – Brittleness. Stamping complex stator geometries with fine teeth is borderline impossible at production scale. Laser cutting works but the heat-affected zone partially crystallizes the material, degrading properties. – B_sat ≈ 1.56 T — workable but lower than NOES. You need a larger core for the same torque, partially negating the loss advantage. For transformer cores and certain axial-flux topologies with simple wound shapes, amorphous makes sense. For the radial-flux frameless torque motors that dominate humanoid joint applications, it’s not practical today. Thin-gauge NOES with bonded stacking gets 80% of the theoretical benefit at 20% of the process difficulty.

Our Manufacturing Process for Humanoid Joint Motor Stacks

1. Material incoming inspection — Electrical steel coil verified via Epstein frame against the mill certificate. We test a sample from each coil before it enters production. Core loss at 1.0 T/400 Hz and magnetic polarization at 2500 A/m are the gate values.
2. Slitting — Coil slit to exact strip width for stator or rotor OD. We handle gauges from 0.10 mm to 0.50 mm.
3. Stamping or laser cutting — For gauges ≥ 0.15 mm and volumes above ~5,000 stacks, high-speed progressive dies at up to 400 SPM (burr spec: ≤ 10 μm at 0.20 mm). For thinner material, prototypes, or fine features (tooth width < 2 mm): fiber laser cutting. Wire EDM available for ultra-precision prototype work.
4. Annealing — Stress-relief anneal at 750°C in N₂ atmosphere for NOES. For Ni-Fe alloys: atmosphere and temperature matched to the specific grade — 850–1000°C in protective gas for 40–42% Ni types; 1100–1200°C in dry H₂ (dew point ≤ −40°C) for high-permeability 49% Ni types. Vacuum anneal available for Co-Fe.
5. Stacking and bonding — Self-bonding varnish (C-3 class per EN 10342) or post-stamp adhesive application. Heat and pressure curing in controlled fixtures. Concentricity verified by laser measurement to ±0.01 mm.

1. Final inspection — Dimensional CMM, burr height measurement, insulation resistance between laminations. For high-spec orders: sample-destructive core loss verification on finished stacks.
2. Skew assembly — If specified, sub-stacks assembled with angular offset per the motor design’s skew schedule.

Prototype lead time: 7–15 working days. Production: 6–8 weeks.

FAQ

If you’re developing joint motors for a humanoid robot program and need lamination stacks that actually hit the cogging and ripple specs your FEA promised, reach out to our engineering team. We’ll review your motor design, recommend a material and stacking approach, and quote prototype through production volumes.