Wearables

A modular elbow exoskeleton designed from 400 synthetic bodies

Jun 1, 20253 min read
Render of the elbow exoskeleton on the average size-M arm
Undergraduate thesis: Aguirre Colorado, Oscar Andrés. Diseño geométrico de exoesqueleto modular y parametrizable para movimientos de rotación articular. Universidad de los Andes, 2025.

Which arm do you design an exoskeleton around? In this project the answer was neither a single user nor an anthropometric table, but a population: 400 synthetic bodies generated with the SMPL-X parametric model and grouped into three sizes with K-Means. The average surface of each size, together with its point-by-point standard deviation, was used as the direct geometric reference for a modular, wearable elbow exoskeleton.

Context

Most exoskeletons are designed around one reference body and then adjusted with straps and stops. The result tends to fit few people well, and the motor axis stops coinciding with the joint axis as soon as the user changes. That misalignment produces parasitic forces on the skin and is one of the reasons wearable devices are uncomfortable.

The starting point here was to treat body shape as a statistical quantity rather than a constant.

From 400 bodies to three sizes

Two hundred male and two hundred female bodies were generated with the SMPL-X add-on for Blender, a model trained on real 3D scans. The left arm and right leg were separated from each body, and all surfaces were aligned on the rotation axis of the joint.

Four hundred overlaid arm surfaces

The 400 left-arm surfaces, aligned on the elbow rotation axis

K-Means (k = 3) on lengths and widths produced groups S, M and L. For each group an average surface was computed, along with the standard deviation of every vertex.

Arm clustering

Arm clustering into three sizes (k = 3)

The deviation map has an immediate design reading: the regions that vary least between people, close to the elbow, are the best places to anchor the device. The deviation within each size is also substantially lower than that of the whole-population average.

Standard deviation on the average size-M arm

Average size-M arm. Color shows the standard deviation of each mesh point; blue marks the most stable regions

Exoskeleton design

Two components were designed on the size-M average surface:

  • A textile sleeve placed over the low-variability region. Its cutting pattern was obtained by unwrapping the mesh in Blender (UV unwrapping) and it was sewn in sports fabric.
  • A modular rigid structure printed in PLA, with upper and lower supports, a motor mount, and fastening elements sewn to the sleeve.

A variable-length unit (ULV) sits between the motor mount and the upper support: a flexible part that stretches and compresses longitudinally without losing stiffness in the other directions. It absorbs the displacement of the elbow axis during flexion, so that the motor stays aligned with the joint.

Variable-length unit

The variable-length unit in contraction and expansion

Results

The size-M prototype was built and tested on the author's arm. It allows the full elbow flexion range (0–150°) and does not block pronation or supination. Parts can be separated and swapped without special tools.

Prototype in use

3D-printed prototype mounted on the textile sleeve

What is missing

Validation was qualitative and involved a single user. The device has no actuation or sensing yet: it is a geometric proposal, not a working exoskeleton. No structural analysis was carried out and sizes S and L were not built, although their reference surfaces are already computed. Being synthetic, the dataset inherits the biases of the scans SMPL-X was trained on.

How it fits in Robiolab

The work tackles, from the geometry side, a problem shared by every wearable device in the lab: the fit between a rigid structure and a body that is soft and different for each person. The ULV is a compliant mechanism in the service of that fit, and the sizing methodology can be applied to other projects in the group, such as the elbow orthosis for brachial plexus injury.

ExoskeletonsAnthropometrySMPL-XClusteringParametric design