Forget S, M, L. Every robot that rolls off the same production line is dimensionally identical. That means robot garment sizing is about platform templates, not size ranges. I've spent three years building those templates, and this guide walks through how the whole system works.
I remember the first time someone asked me for a "medium" for their Tesla Optimus. It was a hotel chain, and the operations manager had spent twenty years ordering staff uniforms. To him, clothing meant picking a size from a dropdown. Fair enough. But robot clothing sizing does not work that way, and understanding why saves a lot of confusion down the line.
Human sizing exists because human bodies vary wildly. Two people who are both 180cm tall might differ by 15cm in shoulder width. The whole S/M/L system, and all its numbered variants, is a statistical compromise. You are picking the bell-curve bucket that is closest to your actual body, then hoping the drape and stretch handle the rest.
Robots do not have this problem. Every Tesla Optimus that comes off the line in Fremont has a 44cm shoulder width. Not approximately. Exactly. The chest circumference is 98cm on every single unit. There is no bell curve because there is no biological variation. This fact changes everything about how you approach robot garment sizing.
Once a garment pattern is built and verified on one unit of a given platform, it fits every unit of that platform with zero alteration. The engineering cost is front-loaded into pattern development. But the per-unit cost of sizing errors drops to literally zero. I have shipped thousands of garments to fleet operators who never had to return a single piece for fit issues. Try that with human uniforms.
I have seen this attempted more times than I can count, and it fails for reasons that go well beyond poor fit. Human clothing assumes a soft substrate. Shoulders that compress. A torso that has give. Arms with a slight natural bend. A robot chassis is rigid aluminum and carbon fiber. Fabric that drapes naturally on a person just pulls and bunches on a robot because there is nothing underneath to absorb the slack.
The joint problem is worse. A human jacket sleeve assumes elbow flexion of about 145-150 degrees. Many robot platforms exceed 180 degrees. When we first fitted a Figure 03 with a standard cotton blend, the shoulder actuator overheated within 40 minutes because the fabric was binding against the housing and blocking the ventilation pathway. The sleeve tore at the elbow crease on day two. Human garments are not engineered for the forces and geometries that humanoid robot measurements demand.
Beyond mechanics, there is the sensor issue. Robots carry LiDAR, cameras, depth sensors, and force sensors at positions all over their body. A human shirt does not have RF-transparent windows over the left pectoral. It does not have mesh panels where the stereo cameras need line of sight. Putting a cotton Oxford on your robot is like putting a blindfold on your employee and expecting them to greet guests.
If you are wondering how to measure a robot for clothing, the honest answer is: you probably should not have to. For the five major production platforms, all humanoid robot measurements are already known down to sub-millimeter precision. We maintain 3D scan datasets for each, and your order just references the platform name.
But if you are working with a custom or modified platform, here is the measurement protocol we follow. It is more involved than measuring a person, because you are measuring a rigid body with moving parts.
These are the dimensions that actually matter for robot garment sizing. I will spare you the full 68-point scan protocol and focus on what drives pattern decisions.
Those measurements give you a static fit. But static fit is only half the picture. You also need articulation data, which brings us to the next step.
This is what separates a robot clothing size guide from a human one. A human tailor asks you to raise your arms and checks the pull. We do something similar, but across every degree of freedom, documented with motion capture.
For each joint, we record the full range of motion and map how the surface geometry changes at every point in that range. The shoulder actuator on a Tesla Optimus, for instance, creates a 2cm protrusion shift when the arm goes from neutral to full overhead raise. A garment that fits perfectly at neutral will bind at full raise unless the pattern accounts for that shift.
The output is what we call an articulation envelope: a 3D model showing every position the robot can occupy, with fabric tension vectors plotted across the entire surface. This data drives panel shapes, seam placement, and fabric selection for each zone.
AVDI does not sell sizes. We sell platform templates. When you order a garment, you select your robot platform, not a size. The garment is produced from a template built specifically for that platform's humanoid robot measurements.
For each platform we support, the process starts with a high-resolution 3D scan at sub-millimeter accuracy. We scan the robot in neutral pose and at twelve key articulation positions. Then we map every degree of freedom, every surface geometry change, every stress point. From that data, our pattern software generates panel shapes and seam routes optimized for the specific geometry. Seams follow low-stress paths. Panels distribute fabric evenly across the articulation range.
Each pattern goes through at minimum three prototype iterations on live hardware. We test for range of motion restriction, sensor occlusion, thermal buildup, charging access, and visual appearance from multiple angles. I have rejected patterns that passed every functional test because the drape looked wrong from the side. Aesthetics matter. A robot in a garment that works perfectly but looks like a trash bag is not a solved problem.
The approved pattern becomes the permanent production template. Every garment ordered for that platform comes from this template. No sizing variation. No grading. One pattern per platform per garment style.
Since Tesla Optimus is our most-ordered platform, here are the exact dimensions that define the Tesla Optimus clothing size template. If you are comparing against other suppliers or trying to understand why a human "medium" does not work, these numbers tell the story.
Optimus stands 173cm with a 44cm shoulder width, 98cm chest, 78cm waist, and an 82cm inseam. Those proportions sit close to an average human male, which is why people assume human clothes will work. They will not. The actuator housings at the shoulder, elbow, and knee change the surface geometry in ways that human patterns cannot account for. And the flat, rigid chest panel creates a completely different drape profile than a human ribcage. Read our full Tesla Optimus platform guide for the engineering specifics.
To show how much platforms differ, and why a universal robot clothing size guide is impossible, here are the key humanoid robot measurements across the five platforms we serve.
| MEASUREMENT | OPTIMUS | FIGURE 03 | XPENG IRON | UNITREE H1 | ATLAS |
|---|---|---|---|---|---|
| Height | 173cm | 170cm | 178cm | 180cm | 150cm |
| Weight | 57kg | 60kg | 70kg | 47kg | 89kg |
| Chest | 98cm | 94cm | 106cm | 88cm | 118cm |
| Shoulder Width | 44cm | 42cm | 48cm | 40cm | 52cm |
| Arm Length | 62cm | 60cm | 66cm | 64cm | 54cm |
| Inseam | 82cm | 78cm | 86cm | 88cm | 64cm |
| DOF | 40 | 42 | 60 | 19 | 28 |
Look at Atlas versus Unitree H1. Atlas is 30cm shorter but has a chest that is 30cm wider. Its shoulder width is 12cm broader despite being a foot shorter. A garment cut for H1 would not even pull over Atlas's head. These are not minor variations. They are completely different body plans that require completely different engineering.
The DOF column is just as important. XPeng Iron has 60 joints to accommodate. Unitree H1 has 19. That difference means Iron garments need articulation panels at 41 additional joint locations. The garment complexity scales with DOF count, which is why the Iron coverall has 38 individually engineered panels compared to 22 for the H1 equivalent.
Not every robot is a production platform with tens of thousands of units. Research labs, startups building proprietary humanoids, and companies with modified factory units all need clothing for platforms that have no existing template.
I worked with a university robotics lab last year that had built a custom upper body on a modified Unitree H1 lower half. The humanoid robot measurements did not match any existing template because the torso was 6cm wider and the shoulder actuators were completely different units. We built a custom template from their CAD files and had a fitted prototype on their robot in four weeks.
For custom platform robot garment sizing, we need either physical access to the robot for scanning or CAD files of the chassis geometry. Articulation range specs save us significant time. The pattern development process follows the same steps as standard platforms, but the cost is higher because the engineering is not amortized across a large customer base. See our custom robot clothing guide for full details on timelines and pricing.
Once a custom template exists, it becomes your platform's permanent template. Reorder pricing drops to standard catalog rates because the engineering is done. You paid for it once; now every future order is just production cost.
No. I have seen people try this with 3D-printed mannequin torsos and modified human patterns. The proportions might approximate correctly, but you will miss every actuator housing, sensor window, ventilation pathway, and charging port. The garment will restrict motion, block sensors, trap heat, and look terrible. Scaling is not sizing.
Manufacturing tolerances on production humanoids are typically under 0.5mm on critical dimensions. That is well within the tolerance of fabric behavior. We have never measured a dimensional difference between two units of the same platform that affected garment fit. The whole point of robot garment sizing is that identical machines get identical clothing.
Aftermarket sensor arrays, custom end-effectors, and structural modifications change the surface geometry. If your robot has been modified, tell us during ordering. Minor modifications (a bolt-on sensor pod, a different gripper) usually require only a pattern adjustment. Major structural changes may require a partial or full re-template. We handle both through the custom order process.
Beautifully. Because every unit is identical, a fleet order of 50 garments means producing 50 copies of the exact same pattern. No size assortment, no exchanges, no fit checks on individual units. Your 50th Optimus wears the same garment as your first. Read our corporate robot uniform guide for fleet-specific details.
Every AVDI garment ships with a 100% fit guarantee. If a garment does not fit your robot platform as specified, we remake it at no cost. This is not a marketing gesture. It is a statement of confidence in the template system. We know the exact humanoid robot measurements of every platform we support, and we test every pattern on live hardware before it enters production.
In three years of shipping, our fit-issue rate is under 0.3%. When issues do occur, it is almost always a modified unit that was not disclosed at ordering. Even then, we work with the operator to fix it.