Problem

People with weak arm muscles, from muscular dystrophy, an injury or simply carrying a child all day, can use powered arm supports. Most are bulky, heavy and obvious. We wanted something a person could wear under a jacket without anyone noticing, that helps with one movement: bending the elbow.

The closest reference was ETH Zurich's Myoshirt, a textile exomuscle with a motor-driven cable that assists the shoulder. Its off-body actuator and control box weighs about 4 kg. We went smaller and simpler: one joint, hobby-grade parts, everything in a bag.

How it works

  1. A handheld joystick tells an Arduino UNO to step a NEMA 17 motor. A software step count limits travel, the default state is off, and a kill switch cuts the supply.
  2. The motor sits on a plate over a padded shoulder brace and turns a 3D-printed spool, which winds in fishing line.
  3. The line comes over the shoulder and down the front of the arm to a band set above the biceps. Lower down, it slipped.
  4. Between that band and a forearm band, the line runs back and forth through sewn-on steel hook-and-eye fasteners, about four strands, and is tied off.
  5. Winding shortens every strand at once. The forearm band is pulled toward the upper band, and the forearm swings up about the elbow, the way the biceps pulls. Unwinding lets the arm lower under its own weight.
Close-up of a denim sleeve with two grey bands; orange fishing line runs in several parallel strands between hooks on the bands.The prototype worn in the lab: shoulder brace with the motor plate, wires down the arm, a forearm band, and a white jerrycan in hand.
Left: iteration 7, the hook-and-line routing across the elbow. Right: the final layout with the motor on the back of the shoulder, holding a jerrycan of unknown mass.

The hooks replaced 3D-printed pulleys with bearings. We chose them because they are hardened steel, smooth enough to act like a bearing, light, easy to sew on and strong. Ten iterations got us there: finger cables, a glove harness, cloth straps, a shoulder-strap platform, bearing pulleys, then hooks, the stepper, a back-mounted base plate and an undersleeve for comfort.

What the forces say

We never instrumented the prototype, so I built a quasi-static model of the elbow with the same hook geometry as the 3D model, standard segment masses for a 70 kg adult and capstan friction at every hook. Every parameter is an estimate, and the code is open. Three findings stand out.

Line chart of force gain against number of strands. The ideal line rises one-for-one; with friction 0.15 four strands give 2.26 and eight give 2.60. A dotted holding bound rises steeply.
Friction caps the tackle. The line slides around each hook instead of rolling, so with a friction coefficient of 0.15 four strands pull 2.26× the line tension, not 4×. Eight strands reach only 2.60.
Two panels of device torque as a percentage of arm weight torque against elbow angle, for 1, 2 and 4 strands. Lifting stays below 100 percent near 90 degrees; holding with four strands rises well above it.
Better at holding than lifting. With the motor torque we assumed (0.2 N·m) and a 20 mm spool, four strands give about 82% of the torque needed to hold the forearm and hand at 90° while lifting. When holding, the same friction works for the device, up to several times that. This matches what we saw: it suits static support more than active movement.
Heat map of liftable hand load at 90 degrees against motor torque and spool radius, with contour lines at 0, 1, 2 and 3 kg. The baseline sits below the 0 kg line.
The spool is as strong a lever as the motor. Halving the spool radius doubles line tension. With 0.4 N·m and a 10 mm spool, four strands could lift about 1.8 kg at 90°, at the cost of speed.

One more correction. Our slides once claimed a 20.3 kg lifting capacity. The formula on that slide actually evaluates to 2.04 kg. It also assumed a gearbox we rejected and gave force at the spool, not at the hand. We withdrew the figure.

Demonstrations

Videos have no sound and play only when you press play.

Threading the hooks · 58 s. Iteration 7, at home: fishing line looped through sewn-on hook-and-eye fasteners on two bands, pulled by hand to bend the elbow. This is where the multi-strand routing started.
Joystick control · 21 s. The final layout in the lab. The joystick drives the stepper at the shoulder, and the line raises and lowers the forearm. Cropped to keep bystanders out of frame.
Holding a jerrycan · 61 s. Raising the arm while holding a jerrycan. Its contents and mass were not recorded, so this shows function, not capacity.
Shoulder unit, close up · 39 s. Picking up the jerrycan, then close-ups of the motor plate on the shoulder brace and the line flexing the arm.

What we showed, and what we did not

ClaimStatus
Raises and lowers the forearm by joystickShown on video
Works while holding a jerrycanShown; jerrycan mass unknown
Can be worn under a jacket with the electronics in a bagShown in a team test; no observer study
Force needed to lift the arm holding an object7.6 kg peak on a luggage scale (team reading). This is the demand on an assist, not what the device produced
User feedbackOne volunteer, one session: the arm moved; she was surprised it carried the jug but said it did not feel strong. The two-button control confused her, so we switched to a joystick
Assist force, efficiency, muscle activityNot measured
System massTarget under 3 kg; never weighed

This is a course prototype, not a medical device, and we make no claim of clinical benefit or novelty. The paper sets out a bench rig and a small user study that would turn the model's predictions into measurements.

Team and credits

  • Sreeram VasanthTeam lead. Market research and roadmap, spool and coupler design, the hook-and-line arm arrangement and its anchoring, kill switch, hardware, wiring and housing. Force model and working paper.
  • Liau Kheng Wee BryanHardware. Printed mounts, housings and pulley-system iterations, safety.
  • Dai JingyangElectronics. Stepper driver wiring, Arduino code, step limit, Bluetooth to motor.
  • Shreya JindalProcurement and budget, Bluetooth app, sensor research.
  • Sarah Tan Jun LeCommunications and team blog, usability session.
  • Dr HoSupervisor.

Thanks to Prof Andy, Han Yang, Jeremy and Dr Ho for their guidance, to Justin of SG Enable for talking us through what users need, and to the volunteer who tried it on.

In memory of Mr Tony, our lab manager, who helped us more than he had to.

Read more