For a lot of humanoid robot shell work, CNC machining is the wrong process. Figure AI said as much publicly when they introduced Figure 03 — the robot is wrapped in soft textiles instead of hard machined parts, and the company’s BotQ manufacturing programme moved deliberately toward injection molding, die casting, metal injection molding, and stamping. Parts that once occupied a CNC machine for over a week now come out of steel tooling in seconds.
That’s the honest starting point. It’s also why this guide exists.
Because the corollary is equally true: there are shells on every humanoid platform where CNC machining is not just appropriate but irreplaceable — and knowing which is which is worth more to your programme than any supplier’s capability list. This guide walks through the four distinct things people mean by “robot shell,” the six processes competing to make them, the volume thresholds where the economics flip, and the technical requirements — thin walls, cosmetic surfaces, sensor cutouts, sealing, safety — that determine whether a shell works.
If you’re sourcing shells for a humanoid programme, the goal here is that you finish this article able to make the machine-or-mold call yourself, on a part-by-part basis, with numbers behind it.
Four Kinds of “Robot Shell” (and Why the Distinction Decides Everything)
Ask five engineers about robot shells and you’ll get five different mental images. One is thinking about the aluminium housing that carries a hip actuator’s bearings. Another is picturing the sculpted thigh panel that gives the robot its silhouette. A third has the head in mind, with its camera windows and microphone ports.
These are genuinely different parts with different requirements, and conflating them is the single most common source of bad manufacturing decisions in humanoid programmes. So before anything else, the taxonomy.
Actuator and Joint Housings: Structural Enclosures Under Load
These are the enclosures that carry motors, reducers, and bearings inside each actuated joint. They’re structural in the strictest sense — they constrain bearing alignment, transmit joint torque into the limb structure, and act as heat sinks for the motor and driver electronics.
Requirements are unambiguous: bearing bores to H6 or better, mounting flange flatness within 0.02 mm, sealing groove geometry that holds an O-ring at consistent compression, and thermal conductivity sufficient to move motor heat into the surrounding structure. Cosmetics are irrelevant — these parts sit under everything else.
We covered these in depth in our guide to robot joint components. For this article, the relevant point is that they belong firmly in the machining category regardless of volume, for reasons we’ll get to.
Structural Chassis and Frame Shells: The Load-Bearing Skeleton
One level out from the joints, you find the shells that make up the robot’s torso, pelvis, and limb structures. These carry load paths between joints, house batteries and compute, and define the robot’s overall structural stiffness.
They’re often large — a humanoid torso shell can be 400 to 600 mm in its longest dimension — and typically thin-walled with integral ribbing. Requirements centre on stiffness-to-mass ratio, dimensional stability, and precise mounting interfaces where joints and subsystems attach. Some are visible; many sit under exterior panels.
Exterior Body Panels: The Parts People See and Touch
These are the cosmetic shells — thigh covers, chest plates, shoulder caps, forearm shrouds. They carry little or no structural load. Their job is to enclose, protect, and define the robot’s visual identity.
This category has a requirement no other shell type has: it has to look right. Surface uniformity, colour consistency across batches, panel gap alignment, and tactile quality all matter — and none of them appear on a conventional engineering drawing. This is also the category most likely to migrate away from machining as volumes climb.
Sensor Housings and Head Shells: Precision Where Perception Depends on It
Head shells, LiDAR shrouds, camera bezels, and microphone arrays sit in a category of their own. They’re cosmetic and precision-critical simultaneously.
A camera window that’s 0.3 mm out of position shifts the camera’s optical axis relative to the robot’s kinematic frame, which either forces recalibration or degrades perception accuracy permanently. Sensor cutout position tolerances on robot head shells typically run ±0.1 to ±0.2 mm — an order of magnitude tighter than a cosmetic panel needs, and beyond what most low-cost forming processes hold reliably.
Quick Reference: Which Shell Type Maps to Which Process
| Shell type | Structural? | Cosmetic? | Precision-critical? | Typical best process |
|---|---|---|---|---|
| Actuator / joint housing | Yes | No | Yes | CNC at all volumes; die cast + machining at very high volume |
| Structural chassis / frame | Yes | Sometimes | Moderate | CNC to ~500 units, then die casting + secondary machining |
| Exterior body panel | No | Yes | No | CNC to ~200 units, then thermoforming or injection molding |
| Sensor / head shell | Partly | Yes | Yes | CNC well past 1,000 units, or molded + CNC secondary |
Keep this table in mind through the rest of the article. Almost every recommendation that follows is really a statement about one of these four rows.
What Changed in 2026: The Industry Moved to Tooled Processes
The humanoid shell conversation shifted substantially over the past two years, and any sourcing strategy built on 2023 assumptions is now out of date.
Figure 03 and the Shift Away from Hard Machined Exteriors
When Figure AI introduced Figure 03 in October 2025, the design brief was explicit: the robot had to be safe around people in domestic environments. The company’s answer was to cover the robot in soft textiles and multi-density foam rather than exposed hard machined parts, with the soft goods designed to be washable and removable without tools.
The specifications that came with it tell a consistent story — 1,676 mm tall, 60 kg, roughly 9% lighter than Figure 02, with the mass reduction coming substantially from the shift away from machined metal exteriors. Every joint surface was designed to eliminate pinch points.
Figure isn’t alone. 1X Technologies ships its NEO home robot in a knitted nylon bodysuit. The industry direction for consumer-facing humanoids is unmistakable: soft, compliant, textile-covered exteriors that read as approachable rather than industrial.
BotQ: What Figure’s Manufacturing Transition Actually Tells You
The more instructive disclosure came from Figure’s BotQ manufacturing programme. To reach production rates, the company moved to tooled processes — injection molding, die casting, metal injection molding, and stamping. Their own framing was blunt: parts that previously took over a week on a CNC machine could be produced in under twenty seconds using complex steel molds. The capital cost of that tooling was high, but the payback came quickly at their projected volumes.
There’s a temptation for a machining company to argue around this. We won’t, because the economics are simply correct. Once a shell design is frozen and annual volumes climb into the thousands, hard tooling wins on per-part cost by a margin that no machining optimisation closes.
What the BotQ story doesn’t say — and what gets lost when it’s quoted as “humanoids don’t need CNC” — is that every one of those molds was validated against machined prototypes, that die-cast and molded shells still require secondary machining on precision features, and that the entire transition presupposes a frozen design. Programmes that aren’t there yet operate under completely different economics.
Soft Goods, Textiles, and Compliant Covers: A New Category of Shell
The rise of textile and foam exteriors creates a genuinely new shell category that didn’t exist in humanoid design five years ago — and it has a knock-on effect on the hard parts underneath.
When the outermost layer is fabric, the rigid shell beneath it stops being a cosmetic part. It no longer needs a Class A surface, colour matching, or a premium tactile finish. What it does need is a smooth, snag-free interface for the soft cover to sit against, secure and repeatable attachment points for removable textiles, and generous edge radii so the fabric doesn’t wear through at contact points.
That’s a meaningful reduction in cosmetic requirement — which changes the machine-or-mold calculus, usually in machining’s favour for mid-volume programmes, because the expensive part of machining a cosmetic shell is the finishing, not the cutting.
What This Means for Your Sourcing Strategy
Three practical takeaways:
- Don’t assume your exterior needs a premium machined finish. If a soft cover is going over it, you’re paying for cosmetics nobody will see.
- Don’t assume tooling is the answer because Figure went that way. Figure is building at a volume most programmes won’t reach for years.
- Do separate your shell BOM by the four categories above before you request a single quote. Different rows want different processes, and a supplier quoting them as one lot will give you the wrong answer on most of them.
The Process Landscape: Six Ways to Make a Robot Shell
Here is the honest field of competing processes — including the four we don’t perform. You can’t make a good machine-or-mold decision without understanding what you’re deciding against.
CNC Machining from Billet
Material is removed from a solid block of aluminium, magnesium, titanium, or plastic until the shell remains. No tooling is required beyond workholding fixtures.
Strengths: Zero tooling cost. Design changes cost nothing but reprogramming time. Achievable tolerances of ±0.02 mm or better on any feature. Full material property retention — a machined 6061-T6 shell has the strength of 6061-T6, unlike a casting. Structural and precision features come free with the process rather than requiring secondary operations.
Weaknesses: Per-part cost stays essentially flat with volume. A large thin-walled torso shell can occupy a 5-axis machine for four to eight hours. Material waste is substantial — a shell may start as a 20 kg billet and finish at 1.5 kg.
Realistic window: Prototypes through low-volume production, plus precision and structural features at any volume.
Die Casting
Molten aluminium or magnesium is injected into a hardened steel die under high pressure.
Strengths: Very low per-part cost at volume. Excellent for complex thin-walled geometry. Fast cycle times measured in seconds.
Weaknesses: Tooling investment is significant and lead times run months. Design changes mean new tooling. Cast material properties are lower than wrought equivalents. Dimensional tolerances are looser than machining, and porosity can appear in thick sections. Critical features almost always require secondary machining.
Realistic window: Structural shells at thousands of units annually with a frozen design.
Injection Molding
Molten thermoplastic is injected into a steel mold. The dominant process for consumer product enclosures.
Strengths: Lowest per-part cost of any process at volume. Excellent surface finish straight from the tool. Clips, ribs, bosses, and cable channels can be molded in directly, cutting assembly part count.
Weaknesses: Tooling costs commonly run from tens of thousands to low hundreds of thousands of dollars, with lead times typically in the range of two to three months. Design changes are expensive. Plastic shells are non-structural — they need metal load paths behind them. Warpage and sink marks require design discipline.
Realistic window: Cosmetic exterior panels at thousands of units annually.
Thermoforming and Vacuum Forming
A heated plastic sheet is drawn over a mold by vacuum.
Strengths: Tooling costs are far lower than injection molding. Excellent for large, shallow, shell-like geometry — exactly the shape of many robot body panels. Good surface quality with no layer lines. Tooling lead times measured in weeks rather than months.
Weaknesses: Limited to relatively simple geometry with adequate draft. Wall thickness varies across the part. Precision features require secondary trimming and machining. Not suitable for structural parts.
Realistic window: Cosmetic panels in the hundreds-to-thousands range — the sweet spot between machining and injection molding.
Vacuum Casting with Urethane
A silicone mold is made from a master pattern, then urethane resin is cast into it under vacuum.
Strengths: Very low tooling cost. Fast turnaround. Good surface reproduction. Can simulate a wide range of production plastics.
Weaknesses: Per-part cost does not improve with volume — it typically stays in the region of one to three hundred dollars per part regardless of quantity. Silicone molds degrade after roughly twenty to fifty pulls. Material properties are inferior to injection-molded thermoplastics.
Realistic window: Ten to a hundred cosmetic parts where appearance matters more than material performance. Notably, the master pattern is usually CNC machined.
Additive Manufacturing: MJF, SLS, and FDM
Parts are built layer by layer from powder or filament.
Strengths: No tooling. Geometric freedom that no other process matches. Fast for very low quantities.
Weaknesses: Surface finish shows layer structure, which telegraphs through paint. Anisotropic material properties. Cost per part doesn’t improve much with quantity. Dimensional accuracy is generally looser than machining.
Realistic window: One to ten units, internal brackets, and fixtures. Above roughly ten cosmetic parts, thermoforming or machining usually produces a better result per dollar.
Side-by-Side Process Comparison
| Process | Tooling cost | Per-part cost trend | Tooling lead time | Achievable tolerance | Surface quality | Structural? |
|---|---|---|---|---|---|---|
| CNC machining | None | Flat | None | ±0.02 mm | Excellent with finishing | Yes |
| Die casting | High | Drops steeply | Months | ±0.1-0.3 mm | Good, needs finishing | Moderate |
| Injection molding | High | Drops steeply | 2-3 months | ±0.1-0.2 mm | Excellent from tool | No |
| Thermoforming | Low-moderate | Drops moderately | Weeks | ±0.5 mm | Good | No |
| Vacuum casting | Low | Flat | Days | ±0.3 mm | Good | No |
| Additive (MJF/SLS) | None | Nearly flat | None | ±0.3 mm | Layer lines visible | No |
The pattern in that table is the whole argument. Two processes have flat per-part cost curves and no tooling — machining and vacuum casting. Everything else trades upfront investment for per-part savings. Your volume determines which side of that trade you belong on.
The Volume Decision: Where CNC Wins and Where It Loses
This is the section most suppliers won’t write. Here are the actual thresholds.
1-50 Units: CNC Is Almost Always Correct
At this volume, tooling amortisation is hopeless. A fifty-thousand-dollar injection mold spread over fifty parts adds a thousand dollars per part before you’ve made anything.
Machining also gives you something more valuable than unit cost at this stage: design mutability. Prototype programmes change constantly. A machined shell can absorb a design revision in a day of reprogramming. A tooled shell cannot absorb it at all.
The one legitimate competitor in this range is additive manufacturing for parts where precision genuinely doesn’t matter — internal brackets, mockup panels, ergonomic study models.
50-500 Units: The Grey Zone Where Analysis Matters
This is where it gets interesting, and where most humanoid programmes currently sit.
For cosmetic exterior panels, thermoforming starts becoming competitive around one to two hundred units. Its tooling costs are low enough to amortise, its lead times are weeks rather than months, and it produces a cohesive surface that takes paint well.
For structural shells and precision housings, machining usually still wins through this entire range. Die casting tooling doesn’t amortise this fast, and cast material properties often can’t meet structural requirements anyway.
The deciding questions:
- Is the design frozen, or will it change in the next six months?
- Does the part carry structural load?
- Does it have features tighter than ±0.2 mm?
- Is a soft cover going over it?
Two or more “yes” answers on the last three, or a “no” on the first, and machining is probably still your answer.
500-5,000 Units: Bridge Tooling and Hybrid Strategies
At this volume, pure machining of cosmetic panels stops making economic sense, but full hard tooling is still a large commitment for a design that may not be final.
Bridge tooling fills the gap — aluminium injection molds instead of hardened steel, thermoform tooling instead of injection molding, or cast-plus-machine hybrid approaches. Bridge tools cost substantially less than production tooling and come faster, at the price of shorter tool life.
The hybrid strategy that works well here: cast or mold the bulk geometry, then machine only the precision features. A die-cast torso shell with CNC-machined joint mounting interfaces and sensor cutouts gets you casting economics on the mass and machining precision where it matters.
5,000+ Units: Hard Tooling Territory
Above roughly five thousand units annually with a frozen design, hard tooling wins decisively on cosmetic and non-structural shells. This is the territory Figure’s BotQ programme was built for, and their reasoning holds.
Machining’s role doesn’t disappear — it shifts. At this volume, machining produces the master patterns, the fixtures, the secondary operations on cast parts, and the precision housings that tooling can’t hold.
The Exceptions: Where CNC Wins at Any Volume
Some parts never leave machining regardless of quantity:
- Bearing seats and any feature requiring H6 or tighter
- Sealing surfaces where groove geometry determines IP rating
- Optical and sensor interfaces with sub-0.2 mm position requirements
- Structural elements where cast material properties are insufficient
- Any part where the design isn’t frozen
Where CNC Machining Is Irreplaceable
Having been honest about where machining loses, here is where it doesn’t — and these categories represent the majority of the machined content on a mature humanoid platform.
Structural Shells That Carry Load
Cast aluminium has lower yield strength than wrought aluminium of the same nominal alloy, and castings can contain porosity that becomes a fatigue initiation site under cyclic loading. A humanoid robot’s structural shells see millions of load cycles over their service life.
For shells where structural failure is a safety event — pelvis structures, torso load paths, limb mounting interfaces — machined billet gives you the full mechanical properties of the specified temper with no porosity risk. That’s not a preference; on a lot of platforms it’s a requirement.
Precision Sensor Windows and Optical Interfaces
Perception systems are calibrated to a kinematic frame. If the camera moves relative to that frame, the calibration is wrong.
Sensor cutout position tolerances of ±0.1 to ±0.2 mm are routine on humanoid head shells, and they need to hold across an entire production batch, not just on the first article. Machining holds this natively. Thermoforming and injection molding generally require a secondary machining operation to achieve it — which means you’re paying for machining anyway, plus the tooling.
Bearing Seats, Sealing Grooves, and Functional Features
Any shell that carries a bearing, seals against ingress, or provides a precision mating interface has features that molding processes cannot hold. Gasket grooves need consistent depth and width to deliver uniform compression — uneven gasket compression is the leading cause of IP-rating failures in the field. Bearing bores need H6 tolerance and controlled surface finish.
On die-cast shells, these features are machined after casting. The machining doesn’t go away; it just gets more expensive because you’re now fixturing a casting with its own dimensional variation.
Master Patterns and Bridge Tooling for Cast and Molded Shells
Vacuum casting requires a master pattern, and that master is almost always CNC machined because it needs to be dimensionally accurate and surface-perfect. Thermoform tooling is machined. Injection mold cavities are machined and EDM’d.
Every tooled process in this article depends on machining somewhere upstream. When a programme “moves away from CNC,” what usually happens is that machining moves from producing parts to producing the means of producing parts.
Secondary Machining on Cast and Molded Shells
The final and largest category. Cast shells need their mounting faces machined flat, their bearing bores bored, their threaded holes tapped, and their sensor cutouts trimmed to position. Molded shells need trimming, precision hole placement, and interface machining.
For a mature humanoid programme running die-cast structural shells, secondary machining is typically twenty to forty percent of the total machining spend — and it continues indefinitely, because every cast part needs it.
Materials for Humanoid Robot Shells
Material choice on a robot shell is a three-way negotiation between mass, stiffness, and finish quality — with manufacturability as the constraint that decides whether your chosen combination is achievable.
6061-T6: The Cosmetic and General-Purpose Default
6061-T6 is the default for a reason. It machines cleanly, it’s dimensionally stable, it’s readily available in plate and billet, and — critically for exterior shells — it takes anodising beautifully and consistently.
Yield strength around 276 MPa is adequate for most non-structural and lightly-structural shells. Density is 2.7 g/cm³. Cost is the lowest of the aluminium options.
For cosmetic panels and head shells where appearance is a requirement, 6061 is usually the correct answer even when a stronger alloy would work structurally, because the finish quality is more consistent.
7075-T6: Structural Shells and the Anodising Trade-Off
7075-T6 roughly doubles 6061’s yield strength — around 503 MPa — at the same density. For structural chassis shells where wall thickness is being minimised, that strength difference translates directly into mass savings.
There’s a catch that catches a lot of design teams, and it’s worth stating plainly: 7075 does not anodise like 6061. Its zinc content tends to produce a yellowish cast under clear anodising and less consistent colour under dyed finishes. For a structural shell hidden under a cover, this is irrelevant. For a visible cosmetic panel, it can mean a batch of parts that don’t match each other.
The practical rule we suggest: 7075 for structure, 6061 for anything the customer sees, and never mix the two alloys within a single visible assembly if colour matching matters.
Magnesium AZ31 and AZ91: 30% Lighter, With Real Handling Requirements
Magnesium is the lightest structural metal available — roughly thirty percent lighter than aluminium at 1.74 to 1.81 g/cm³ depending on alloy. On a humanoid platform, where mass in the limbs amplifies through every joint’s torque requirement, that reduction has outsized value.
The common wrought grades for machined shells are AZ31 (good formability and weldability, moderate strength) and AZ91 (higher strength, more common in cast form). Magnesium also damps vibration better than aluminium and machines quickly, with lower cutting forces than aluminium at comparable material removal rates.
The reason more suppliers don’t offer it is safety. Magnesium chips are combustible, and fine magnesium dust is a genuine fire and explosion hazard. Machining magnesium safely requires appropriate coolant selection — water-based coolants react with magnesium to release hydrogen — dedicated chip handling and containment, Class D extinguishing capability at the machine, and operator training that most general job shops don’t have.
Richconn machines magnesium robot shell components in production. If you’re considering magnesium for a shell programme, the questions worth asking any supplier are what coolant they run, how they handle and store chips, and whether magnesium work shares a cell with steel or aluminium — the answers tell you quickly whether the capability is real.
Titanium Ti-6Al-4V: Where Mass and Strength Both Matter
Ti-6Al-4V sits at 4.43 g/cm³ with yield strength around 880 MPa — heavier than aluminium but with a substantially better strength-to-mass ratio and excellent fatigue performance.
On humanoid shells, titanium appears selectively: high-load structural brackets within a shell assembly, mounting interfaces that see impact loading, and medical or laboratory robots where sterilisation compatibility matters. Whole titanium shells are rare because the machining cost premium — typically three to four times aluminium for equivalent geometry — is hard to justify across a large surface area.
Titanium CNC machining has particular requirements around cutting speed, coolant delivery, and chip control that are worth discussing before you commit a design to it.
Engineering Plastics: ABS, PC, PEEK, and Machined Polymer Shells
Machining plastic shells is unusual but occasionally correct — typically when you need a small number of parts in the exact production resin to validate fit, appearance, or optical performance before committing to tooling.
- ABS — takes primer and paint well, machines cleanly, the standard choice for painted cosmetic prototypes
- Polycarbonate — essential for transparent windows and sensor covers; machines and polishes to near-optical clarity
- PEEK — high strength and temperature capability, used where a polymer shell needs to be structural or sterilisable
- Acetal (POM) — dimensionally stable and machines beautifully, but paint adheres poorly; avoid for painted cosmetic parts
One design note that saves grief: if your production shell will be injection molded in ABS, machine your prototype in ABS rather than acrylic or acetal. The paint and finishing behaviour will match, and you’ll catch appearance problems before tooling.
Material Selection Matrix by Shell Type
| Shell type | First choice | Alternative | Avoid |
|---|---|---|---|
| Actuator / joint housing | 6061-T6 | 7075-T6, magnesium | Plastics |
| Structural chassis | 7075-T6 | Magnesium AZ31, 6061-T6 | Cast alloys at low volume |
| Exterior panel (visible) | 6061-T6 | ABS, PC | 7075 (colour inconsistency) |
| Exterior panel (under soft cover) | 6061-T6, magnesium | ABS | Titanium (unnecessary cost) |
| Head / sensor shell | 6061-T6 | PC for windows | Acetal (paint adhesion) |
| Weight-critical limb shell | Magnesium AZ31 | 7075-T6 | Steel |
Thin-Wall Machining: The Core Technical Challenge
If there’s one technical problem that defines robot shell machining, this is it. Almost every shell on a humanoid platform is a large, thin-walled, curved structure — precisely the geometry most prone to distortion.
Why Robot Shells Are Deformation-Prone by Nature
A wall is generally considered thin below about 2 mm. Robot shells routinely specify 1.5 to 2.5 mm walls across surfaces measuring hundreds of millimetres, which puts nearly every shell in the difficult category by definition.
Three mechanisms cause the trouble. Cutting force deflection — a thin wall behaves like a cantilever, flexing away from the tool and springing back after it passes, leaving walls thicker at the top than the base. Thermal distortion — thin sections have little thermal mass, so localised heating produces localised expansion, and aluminium’s high coefficient of thermal expansion amplifies it. Residual stress release — stresses locked into the billet during rolling or extrusion release as material is removed, and the part moves.
The third mechanism is the one that surprises people, because the part measures correctly on the machine and then warps overnight.
We’ve written about this problem in more depth in our guide to preventing thin-walled part deformation, which covers the general principles. What follows is how those principles apply specifically to robot shells.
Residual Stress and Pre-Machining Stress Relief
For any shell removing more than roughly seventy percent of the starting billet — which describes most of them — we recommend stress relief before finish machining.
The practical sequence: rough the part leaving 1.5 to 2 mm of stock on all surfaces, stress-relieve thermally, allow the part to stabilise, then finish machine. On large torso shells the difference between doing this and not doing it is often the difference between parts that hold ±0.1 mm on profile and parts that don’t hold ±0.5 mm.
Material selection matters here too. Stress-relieved plate stock behaves far more predictably than extrusion for large shells, and it’s worth specifying rather than leaving to the supplier’s discretion.
Fixturing Strategy: Vacuum Chucks, Soft Jaws, and Sacrificial Support
Clamping force is itself a distortion source. A thin shell clamped hard in a conventional vice will machine correctly and then spring out of tolerance when released.
The strategies that work on robot shells:
- Vacuum fixturing for parts with a flat or near-flat face, distributing holding force across the full contact area
- Custom soft jaws machined to match the part contour — a small tooling investment that pays back immediately on a batch
- Sacrificial support ribs left in the design and removed in a final operation, keeping thin regions rigid through the aggressive cuts
- Low-melt support wax or castable support for complex curved shells with no good clamping geometry
- Sequential release — machining and releasing regions progressively rather than removing all clamping at once
On large curved shells, custom fixturing isn’t optional. Budget for it in the first article cost.
Toolpath and Cutting Parameter Strategy for Thin Walls
The general principle is to keep cutting force low and constant rather than to cut slowly.
- Trochoidal roughing with radial engagement of five to ten percent of tool diameter keeps forces predictable and lets you use the full flute length
- Symmetric material removal — alternate sides rather than finishing one face fully before starting the other, which balances stress release
- Multiple light finishing passes rather than a single heavy one, with the last pass taking 0.1 to 0.2 mm
- Climb milling throughout finishing for consistent chip load and better surface finish
- Sharp tooling — a worn tool dramatically increases cutting pressure, which thin walls translate directly into deflection
- Consistent thermal environment, ideally within ±1°C during finishing on precision shells
Achievable Wall Thickness by Material
| Material | Comfortable minimum | Achievable with careful process | Notes |
|---|---|---|---|
| 6061-T6 aluminium | 1.5 mm | 0.8 mm | Most predictable; best general choice |
| 7075-T6 aluminium | 1.2 mm | 0.7 mm | Higher stiffness helps; more residual stress |
| Magnesium AZ31 | 1.5 mm | 1.0 mm | Low cutting force helps; handling constraints apply |
| Ti-6Al-4V | 1.0 mm | 0.6 mm | High stiffness, but heat management is critical |
| ABS / PC | 2.0 mm | 1.5 mm | Thermal softening is the limit, not force |
| PEEK | 1.5 mm | 1.0 mm | Machines well but relieves stress unpredictably |
These figures assume reasonable rib support and aspect ratios. An unsupported 0.8 mm wall spanning 300 mm is a different proposition from one spanning 50 mm — if your design pushes both thickness and span, bring it to a design review early.
Cosmetic Requirements: What Makes Exterior Shells Different
Internal components are judged by a drawing. Exterior shells are judged by eye — and the requirements that follow from that don’t appear on most engineering drawings at all.
Toolpath Marks and the Path to a Uniform Surface
An as-machined aluminium surface carries visible toolpath marks. Standard machining typically produces around Ra 3.2 μm, and even a fine finishing pass leaves a directional pattern the eye picks up immediately under raking light.
For cosmetic shells, the finishing chain matters as much as the machining. The typical route is a fine finishing pass with consistent stepover and no dwell marks, followed by bead blasting to homogenise the surface, followed by anodising or painting.
The machining-side requirement is consistency rather than pure smoothness — a surface with uniform Ra 1.6 μm blasts to a better result than one alternating between Ra 0.8 and Ra 3.2 μm across different regions, because blasting reveals variation rather than hiding it. Toolpath transitions, stepover changes, and tool changes mid-surface all show up after finishing.
Bead Blasting: Media, Pressure, and Batch Consistency
Bead blasting creates the uniform matte texture that reads as “product” rather than “part.” It’s inexpensive — typically a few dollars per part with sub-day turnaround — and it’s the standard first step for almost every consumer-facing aluminium enclosure.
The variables that determine the result are media type and size, blast pressure, nozzle distance and angle, and dwell time. Consistency across a batch depends on controlling all four, which is why blasting robot shells freehand produces visibly variable results and why fixtured or robotic blasting is preferable for cosmetic work.
Richconn coordinates surface finishing through vetted external finishing partners rather than performing it in-house. In practice that means we specify and control the finish requirement, manage the parts through the finishing process, and inspect on return — but we’re transparent that the blasting and anodising themselves happen at partner facilities. For cosmetic shell programmes we recommend agreeing finish standards with a physical reference sample before production starts, which is worth doing regardless of who performs the work.
Anodising: Type II vs. Type III, and Why 7075 Turns Yellow
Type II (decorative) anodising produces a coating typically 5 to 25 μm thick, accepts dye readily across a wide colour range, and is the standard for cosmetic shells.
Type III (hard) anodising produces 25 to 75 μm of much harder coating with superior wear resistance, but colour options are limited and the finish is typically darker and less uniform. It’s a functional finish, not a cosmetic one.
The dimensional consequence is the same trap that catches designers on joint housings: anodising grows the surface, with roughly half the coating thickness building outward and half penetrating the substrate. On a Type III coating that’s a meaningful dimensional change. Any precision feature on a cosmetic shell — sensor cutouts, bearing interfaces, mating surfaces — either needs masking or needs the coating allowance designed in.
And to repeat the alloy point because it causes real production problems: 7075’s zinc content produces inconsistent colour under anodising, tending yellow under clear. If your visible shells are 7075, either accept the appearance or plan on paint instead.
Colour Consistency Across Production Batches
This is the requirement most commonly discovered too late.
Anodised colour depends on alloy composition, surface preparation, bath chemistry, temperature, and dwell time. Parts anodised in different batches — or from different material lots — can differ visibly even when every parameter is nominally within specification. On a robot with eight visible panels that all need to match, that’s a rejection risk on every batch.
What controls it in practice:
- Specify a single alloy and temper for all colour-matched parts
- Source material from a single lot where possible for a production batch
- Anodise all colour-matched parts from an assembly in the same bath run
- Establish approved physical reference samples — an upper and lower acceptance limit, not a single target
- Define the inspection condition, since colour reads differently under different lighting
Parting Lines, Panel Gaps, and Assembly Fit Tolerances
Panel gaps are the most visible quality signal on a finished robot. A consistent 1.5 mm gap reads as intentional; a gap varying between 1.0 and 2.5 mm reads as poorly made, even if every part is within its individual drawing tolerance.
Gap consistency is a tolerance stack-up problem, and it’s solved the same way as any other: functional GD&T referenced to the mounting datums rather than accumulated size tolerances, with the gap itself specified as a controlled dimension rather than left to fall out of the stack. For cosmetic assemblies we’d suggest specifying panel gap and flush conditions explicitly on the assembly drawing, since it’s the only way a supplier can machine to it deliberately.
Functional Shell Requirements Beyond Appearance
Beneath the cosmetics, robot shells carry a set of functional requirements that drive real machining specifications.
Sensor Cutout Position Tolerance and Perception Calibration
A humanoid head shell may carry ten to twenty sensor apertures — stereo cameras, depth sensors, LiDAR windows, microphone ports, status indicators. Position tolerances typically run ±0.1 to ±0.2 mm, referenced to the shell’s mounting datum rather than to shell edges.
That datum choice matters more than the tolerance number. If cutouts are dimensioned from a shell edge that itself varies, the tolerance is meaningless in assembly. Referencing them to the mounting interface — the features that locate the shell relative to the robot’s kinematic frame — is what makes the specification achievable and useful.
Where a sensor requires optical clarity through a window, the window aperture also needs a controlled surface finish on its sealing land and a defined edge condition, since a burr or chip at the aperture edge can scatter light into the sensor.
IP Rating: Gasket Groove Geometry and Sealing Surfaces
Most humanoid platforms target IP54 for indoor operation, with outdoor or washdown-capable variants reaching IP65 or higher. The rating is determined less by the seal material than by the groove that holds it.
The leading cause of IP failures in the field is uneven gasket compression. What produces even compression:
- A tongue-and-groove or stepped joint geometry that creates a controlled sealing path
- Groove depth and width held to consistent tolerance around the full perimeter — typically ±0.05 mm on depth
- Sealing land surface finish around Ra 1.6 μm or better, with no machining witness marks crossing the seal path
- Fastener spacing close enough to keep compression uniform between fixing points
- Corner radii in the groove generous enough that the gasket doesn’t bridge
Form-in-place gasket dispensing is worth considering for complex shell perimeters, since it follows contours that a moulded gasket can’t. It requires a dispensing path with consistent groove geometry, which is a machining specification.
EMI Shielding and the Anodising Conflict
Robot shells frequently double as EMI enclosures, containing emissions from motor drivers and protecting sensor electronics. Conductive gaskets can achieve substantial attenuation across the relevant frequency range — but only with genuine electrical contact between the gasket and the housing.
Here’s the conflict that catches designs regularly: anodising is an electrical insulator. A shell specified “anodise all over” and also expected to provide EMI grounding will fail EMC testing, and the failure is expensive because it’s usually discovered late.
The fix is straightforward once you know to look for it — either mask the grounding and gasket contact surfaces before anodising, or specify chromate conversion coating (Alodine) on those surfaces instead, which maintains conductivity while providing corrosion protection. Either approach needs to be on the drawing from the start, since retrofitting it means stripping and refinishing parts.
There’s a second-order issue worth flagging: conductive gaskets contain metallic fillers, and dissimilar metal contact in the presence of moisture accelerates galvanic corrosion. On magnesium shells particularly, filler selection matters beyond shielding performance.
Thermal Management Through Shell Design
Humanoid robots dissipate substantial heat from actuators, drivers, and compute. Shells participate in that thermal path whether or not they were designed to.
Machined aluminium and magnesium shells conduct heat well, which lets a shell act as a distributed heat sink if it’s coupled properly to the heat source. Design features that support this: machined thermal interface pads with controlled flatness where components mount, integral fin geometry on internal surfaces, and material continuity in the conduction path rather than thermal breaks at joints.
The counterpoint is that a shell conducting heat well also gets warm to the touch. On robots operating around people — increasingly the design case — external surface temperature becomes a safety specification, which may argue for a thermal break and a separate cooling path rather than conducting heat to the exterior.
Human-Safety Features: Edge Radii, Pinch-Point Elimination, and Soft-Cover Interfaces
Collaborative and domestic humanoids are subject to contact-safety requirements, and shell geometry is where a lot of those requirements get satisfied.
The machining-relevant features:
- Edge radii on all accessible edges, typically 2 mm minimum, with larger radii on surfaces likely to contact a person
- Pinch-point elimination at joint covers — shells that move relative to each other need geometry that closes gaps progressively rather than creating a shear
- Soft-cover interfaces — where textile or foam covers attach, the underlying shell needs snag-free surfaces, secure attachment features, and edge conditions that don’t abrade the cover
- Fastener recessing so no head protrudes above a contactable surface
- Deburring specification treated as a functional requirement rather than a cosmetic afterthought
If your platform is following the Figure 03 direction toward soft exteriors, the shell underneath is a substrate for that cover — and its requirements shift accordingly, usually becoming less demanding cosmetically and more demanding on edge condition and attachment feature precision.
Design for Manufacturability: Robot Shell DFM Guidance
The highest-value input a manufacturing partner gives you happens before the quote. These are the points we raise most often on shell designs.
Minimum Wall Thickness Targets by Material and Process
Design to the process you’ll eventually use, not just the one you’re prototyping with. The gap between them causes expensive surprises:
| Material | CNC minimum | Die casting minimum | Injection molding minimum |
|---|---|---|---|
| Aluminium | 0.8-1.5 mm | 1.5-2.5 mm | — |
| Magnesium | 1.0-1.5 mm | 1.0-2.0 mm | — |
| ABS / PC | 1.5-2.0 mm | — | 1.5-3.0 mm |
If you design a 1.0 mm aluminium wall because machining allows it and later move to die casting, you’ll be redesigning. Designing to the more restrictive process from the outset costs a little mass and saves a redesign cycle.
Designing a Shell That Can Migrate from CNC to Tooling
This is the single highest-value DFM principle in this article, and almost nobody designs for it.
Most humanoid programmes start with machined shells and intend to move to tooling later. But a shell designed purely for machining frequently can’t be tooled without substantial redesign — which means the transition costs a full redesign cycle plus revalidation, right at the moment the programme is trying to scale.
What a migration-ready shell looks like:
- Draft angles designed in from the start, even though machining doesn’t need them
- Uniform wall thickness, since casting and molding both dislike thickness variation while machining is indifferent to it
- No undercuts unless you’re prepared to pay for side actions in tooling
- Ribbing instead of thick sections for stiffness, which suits every process
- Precision features isolated to machinable regions, so they can become secondary operations on a cast part rather than requiring the whole part to be machined
- Radii rather than sharp internal corners, which tooling requires and machining prefers anyway
The cost of designing this way is a small mass penalty and slightly constrained geometric freedom. The benefit is that the same design carries from prototype through production without a redesign — which on a humanoid programme is typically worth months.
Draft Angles: Why You Should Add Them Before You Need Them
Worth its own note because it’s the most-ignored item on the list above.
Machining doesn’t need draft. A 90° wall is as easy to machine as a 92° wall. So machined prototypes routinely have zero draft — and then the shell can’t be cast or molded without adding it, which changes every mating dimension.
Add one to two degrees of draft on all vertical faces at the initial design stage. It costs nothing in machining, it’s invisible in the finished product, and it preserves your tooling option.
Part Consolidation vs. Panel Splitting
Machining and tooling pull in opposite directions here, which is why the decision needs making deliberately.
Machining favours consolidation — a single 5-axis machined shell replacing four fastened pieces eliminates tolerance stack-up, reduces assembly labour, and improves stiffness. Setup cost dominates, so one complex part usually beats four simple ones.
Tooling favours splitting — smaller parts mean smaller, cheaper molds, faster cycles, and simpler tool geometry.
For a programme intending to migrate, the reasonable middle path is to split along lines that make sense for tooling but machine the split parts as a matched set on the same fixture. You get most of the assembly benefit of consolidation while keeping the tooling path open.
Fastener and Insert Strategy for Thin Shells
Thin-walled shells can’t reliably carry tapped threads. A 1.5 mm aluminium wall gives you barely more than one thread engagement on an M3 fastener, and any repeated disassembly will strip it.
The options, roughly in order of preference:
- Threaded inserts — helical or press-in, in a machined boss with adequate wall thickness around it
- Local bosses — thicken the wall locally to 3-4 mm at fastener locations rather than thickening the whole shell
- Through-bolting to a backing structure rather than threading the shell itself
- Captive nuts or nutplates where frequent service access is expected
And a specification point that’s easy to miss: define torque values for shell fasteners. Thin shells deform under over-torque, and panel gap consistency suffers immediately.
When to Bring Manufacturing into the Design Conversation
Earlier than most programmes do.
Manufacturing input during CAD design routinely saves twenty to forty percent of production cost through wall thickness rationalisation, part consolidation, feature simplification, and material substitution. The same input after design freeze typically saves under ten percent, because the remaining changes ripple into neighbouring components.
The point of maximum value is when the shell’s external geometry and mounting interfaces are settled but internal detail is still fluid. That’s when draft angles, wall thickness strategy, and migration-readiness can still be designed in for free.
Planning the Transition from Prototype to Production
Pulling the preceding sections into an executable sequence.
Stage 1: Design Validation (1-20 Units)
Process: CNC machining, with additive for non-critical internal parts.
Objective: Validate fit, assembly, thermal behaviour, sensor alignment, and appearance. Expect multiple revisions.
What to get right: Don’t over-invest in finishing at this stage — bead blast only what you’re showing to stakeholders. Do apply the migration-ready design rules now, because retrofitting them later is expensive. Machine in the material you intend to produce in, so thermal and structural behaviour translate.
Typical timeline: One to three weeks per iteration.
Stage 2: Pilot Production (20-200 Units)
Process: Still predominantly CNC. Vacuum casting becomes viable for cosmetic panels if appearance matters more than material properties.
Objective: Field testing, customer demonstrations, manufacturing process validation. Design should be stabilising.
What to get right: This is when to invest in custom fixturing, since it amortises across the batch. Establish cosmetic reference samples and a finish specification. Start the tooling conversation even though you’re not tooling yet — quotes and lead times inform your Stage 3 timing.
Typical timeline: Four to eight weeks per batch.
Stage 3: Bridge Tooling (200-2,000 Units)
Process: Thermoforming for cosmetic panels, bridge tooling for higher-volume parts, CNC retained for structural and precision shells.
Objective: Reduce per-part cost without committing to hard tooling on a design that may still change.
What to get right: Split the BOM deliberately using the four-category framework. Don’t tool everything — tool the parts with the highest volume and lowest change risk. Keep precision features as secondary machining operations.
Typical timeline: Three to eight weeks tooling lead, then weeks per batch.
Stage 4: Hard Tooling (2,000+ Units)
Process: Die casting and injection molding for bulk geometry, CNC for secondary operations and precision housings.
Objective: Minimum per-part cost at volume.
What to get right: Design must be genuinely frozen. Budget for secondary machining — it doesn’t go away. Maintain a machining supply relationship for the parts that never leave machining, and for the inevitable running changes.
Typical timeline: Two to four months tooling lead.
Cost Modelling: Building Your Own Breakeven Analysis
The calculation is simple enough to do yourself, and worth doing per part rather than per programme:
Machining total = (machined cost per part × quantity) + fixture cost
Tooling total = tooling cost + (tooled cost per part × quantity) + secondary machining cost × quantity
Set them equal, solve for quantity, and you have your breakeven. Then adjust for the factors the arithmetic misses:
- Design change probability — every revision after tooling means new tooling
- Time to market — tooling lead time may cost more in delay than it saves in unit cost
- Cash flow — tooling is capital up front; machining is expense as you go
- Volume confidence — breakeven at 3,000 units is irrelevant if you might only build 1,200
In our experience the arithmetic breakeven and the correct decision differ more often than not, and it’s usually the design change risk that drives the difference.
Common Transition Mistakes
- Tooling too early, before the design is genuinely frozen, then paying for tooling revisions
- Tooling too late, staying on machining well past the point where it costs real money
- Tooling everything rather than splitting the BOM by shell category
- Forgetting secondary machining in the tooled-part cost model
- Designing without draft, forcing a redesign at transition
- Switching suppliers at transition and losing the process knowledge accumulated during prototyping
Choosing a Manufacturing Partner for Robot Shells
Capability Checklist for Shell Manufacturing Suppliers
What actually matters for humanoid shell work:
- 5-axis machining with adequate envelope for large torso and limb shells
- Demonstrated thin-wall capability — ask specifically about their approach to residual stress and fixturing, not just their tolerance claim
- Custom fixture design and fabrication in-house, since off-the-shelf workholding won’t hold curved shells
- Broad material capability across aluminium alloys, magnesium, titanium, and engineering plastics
- Magnesium handling competence if lightweighting matters — ask about coolant, chip handling, and cell segregation
- Managed surface finishing with documented finish specifications and reference sample control, whether in-house or through partners
- CMM inspection with the capacity to verify large-part profile tolerances, not just point dimensions
- DFM engagement during quoting rather than quote-only service
- Honest process guidance — a supplier that never suggests you should tool a part is optimising for their revenue, not your programme
Richconn’s Approach to Humanoid Robot Shell Production
At Richconn, we machine humanoid robot shells across the four categories described in this article — actuator and joint housings, structural chassis shells, exterior panels, and precision sensor housings.
Our capability set for this work centres on 5-axis machining for large curved shells, with in-house custom fixture design because shell geometry rarely suits standard workholding. Material capability spans 6061-T6 and 7075-T6 aluminium, magnesium, Ti-6Al-4V titanium, and machinable engineering plastics including ABS, polycarbonate, and PEEK.
On magnesium specifically: we machine magnesium robot shell components in production, and we’re happy to discuss our handling and coolant protocols in detail with programmes considering it. Given the safety requirements, we’d encourage you to ask those questions of any supplier quoting magnesium work.
On surface finishing, we want to be straightforward about how this works: bead blasting, anodising, and painting are performed by vetted external finishing partners rather than in-house. What we provide is specification control, reference sample management, parts management through the finishing process, and inspection on return. For cosmetic shell programmes we’ll work with you to establish approved reference samples before production begins — which is the step that prevents most batch-consistency problems regardless of where the finishing happens.
We also perform secondary machining on cast and molded shells, which is where a lot of shell programmes end up as they scale. If your programme is heading toward tooling, that transition doesn’t have to mean changing machining suppliers.
How to Start: NDA, Design Review, and Quotation
- Initial contact via richconn.com with a brief description of your programme, target volumes, and timeline
- Mutual NDA — standard practice for pre-commercial robotics work
- Design review covering shell categorisation, material recommendations, thin-wall strategy, migration-readiness, and honest process guidance including where we’d suggest tooling instead
- Quotation with clear pricing, lead time, and fixture cost broken out
- First article with full inspection reporting and, for cosmetic parts, finish reference approval
- Production planning as volumes scale, including secondary machining strategy if you’re moving to tooling
Conclusion
The honest answer to “should I CNC machine my humanoid robot shells?” is: some of them, at some volumes, for specific reasons.
Machine your structural shells, because cast material properties and porosity risk aren’t acceptable in load paths. Machine your sensor and head shells, because ±0.15 mm cutout position isn’t achievable any other way at reasonable cost. Machine everything while your design is still moving, because tooling a design that changes is the most expensive mistake in this entire process. And machine your master patterns and secondary operations indefinitely, because every tooled process depends on machining somewhere.
Tool your cosmetic panels once the design is frozen and volumes justify it. Figure’s transition was the right call for Figure, and the same arithmetic will eventually apply to your programme too — just probably not yet, and probably not for every part.
The manufacturing partner worth having is the one that tells you which is which. At Richconn we’d rather advise a customer to tool a part and keep their machining business on the parts that genuinely need it than machine everything and lose the relationship when the economics catch up.
