Honestly, most robot clothing fabric choices I see are terrible. People grab whatever stretch material is handy and assume it will work. It will not. The relationship between textile and actuator is the core engineering problem. Get it wrong and the fabric fights the machine. Get it right and it disappears into the movement.
Here is what happens when a human bends their elbow: skin stretches on the outside, compresses on the inside, and the soft tissue underneath absorbs the geometry change. Fabric draped over a human arm has a compliant substrate that helps it redistribute tension. The garment industry has had centuries to optimize for this.
A robot elbow is a completely different animal. The actuator housing rotates, creating a hard discontinuity in the surface. No compression. No soft-tissue buffer. The outer surface of the joint extends while the inner surface either gaps or slams together. Any robot joint fabric placed over this mechanism needs to handle these geometry changes without restricting the actuator's torque output, without creating pressure points on the chassis coating, and without looking like a wrinkled mess when the joint returns to neutral.
The naive solution is stretchy fabric. I have tested dozens of stretch materials on servo joints. The problem is always the same. A 4-way stretch textile will accommodate the movement just fine during articulation. But when the joint returns to neutral, the fabric sags. It bags. After eight hours of continuous operation, the garment looks like it has been worn for three months. The recovery rate of standard stretch fabrics just is not fast enough for the cycle counts robots put them through.
The real solution requires matching specific fabric properties to specific actuator types and understanding the mechanical forces at each joint. That is what humanoid robot textile engineering actually involves.
Every actuator type creates a different surface geometry change, and each demands different properties from the robot clothing fabric covering it. Here is what we deal with across the major platforms.
Found at most major joints, including the shoulder, elbow, hip, and knee, on every platform we serve. A rotary actuator creates a pivot point where the surface changes radially. The fabric has to rotate around this pivot without bunching on one side or pulling from the other.
We solve this with a gusset panel centered on the actuator axis. The gusset is cut from a warp-knit textile that stretches in the direction of rotation but stays dimensionally stable perpendicular to it. Critical detail: the gusset anchors to the garment body at four discrete points rather than along a continuous seam. This lets it float freely during rotation instead of dragging the adjacent panels along. I spent about six months getting this anchor geometry right for the Optimus shoulder, and it was worth every prototype iteration.
Used in spine and neck assemblies, particularly on XPeng Iron and Figure 03. Linear actuators create a telescoping surface change where the distance between two attachment points varies as the actuator extends or retracts. The challenge is accommodating length change without creating wrinkles when the actuator is compressed.
We use an accordion-pleat panel here, hidden within the garment's design lines. The pleat folds cleanly when compressed and extends without resistance when the actuator pushes out. The trick is matching the pleat depth to the actuator's stroke length. Too shallow and the pleat bottoms out. Too deep and you get visible bulk in the garment's profile. Iron's spine has a 4.2cm stroke, so we run 7mm-deep pleats. Figure 03's neck actuator has a 2.8cm stroke with shallower 5mm pleats.
Found in hands and some wrist assemblies. Cable drives create subtle, distributed surface movements. The bigger concern is not accommodating the motion but keeping fabric from interfering with cable routing that runs just under the chassis surface. On Figure 03, the Bowden cables for the 16-DOF hand system sit as close as 1.5mm below the forearm shell.
We use rigid standoff panels at cable routing paths. These are thin PETG inserts bonded to the garment's inner face that hold the robot clothing fabric 2-3mm clear of the surface. Zero friction on the cables, no added visual bulk. The standoffs weigh about 8 grams total per sleeve.
Found in high-torque joints, especially on Boston Atlas. Harmonic drives create very little surface geometry change but they generate serious heat. Surface temperatures at Atlas's shoulder harmonic drive hit 62 degrees Celsius during sustained heavy lifting. The robot joint fabric concern here shifts from movement accommodation to thermal management.
We use a thermally isolated standoff lining at harmonic drive locations. The lining holds the outer fabric 6mm off the chassis surface, creating an air gap that both insulates the fabric from heat damage and provides a convection channel for cooling. The lining material is a PA66-based mesh rated to 120 degrees Celsius continuous exposure.
AVDI garments do not look like the inside of a space suit, but structurally they are closer to aerospace composites than to a pair of trousers from a department store. Rather than building from a few large fabric pieces the way a human tailor would, we use 15 to 40 individual panels per garment. Each panel is engineered for the specific surface and movement behavior of the area it covers.
Here is how the zones break down:
Seam placement between panels follows the paths of least stress during movement. This is the opposite of human tailoring, where seam placement follows aesthetic and drape considerations. In humanoid robot textile engineering, the articulation envelope dictates where every seam goes. We optimize seam routing computationally, then adjust by hand after prototype testing on live hardware.
People ask me "what fabric works on robots?" like there is a single answer. There is not. The right servo compatible fabric depends on the zone, the platform, and the operating environment. But here are the baseline criteria every fabric in our library has to meet.
Lab testing tells you what a fabric can survive. Live hardware testing tells you whether it actually works. Every new fabric-actuator combination goes through our rig protocol before it touches a production garment.
The test rig cycles the actuator through its full range of motion 10,000 times with the servo compatible fabric panel installed. Sensors measure fabric tension, actuator power draw, surface temperature at the fabric-chassis interface, and panel position drift. Two hard failure criteria:
This testing runs independently for every platform. A panel design that passes on Optimus's elbow might fail on Figure 03's elbow because the actuator type, housing geometry, and range of motion are all different. There are no shortcuts in humanoid robot textile engineering.
Each platform presents its own problems. A few that keep me up at night:
Atlas shoulder: The harmonic drive runs hot, the joint range exceeds 180 degrees in the sagittal plane, and the actuator housing has a compound-curve profile that changes shape during rotation. I went through nine fabric samples before finding a warp-knit Dyneema blend that handled all three constraints simultaneously. It is the most expensive robot joint fabric in our library by weight, and it is worth it.
XPeng Iron spine: Three linear actuators in series, each with a different stroke length, overlaid with rotational DOF at each vertebral segment. The fabric has to accommodate compound movements where extension, compression, and rotation happen simultaneously. Our solution uses a segmented panel system where each spine segment gets its own independent pleated section. It was a pain to develop but it articulates beautifully.
Unitree H1 knee: Not complex in terms of joint type, but the H1 runs at 3.3 m/s. At that speed, any loose fabric at the knee creates aerodynamic drag and audible flapping. The servo compatible fabric here has to be compression-fit, bonded-seam, zero-float. We essentially laminate the fabric to a neoprene backing at the H1 knee zone. Read the Unitree H1 platform guide for more on the speed challenge.
Every AVDI garment is tested on live hardware before production.
BROWSE THE CATALOG