← Cycloidal Arm

One joint, 21 to 1

Each of the arm's three joints is the same printed cycloidal reducer on a NEMA 17: 22 steel pins, two 21-lobe discs and an eccentric cam. How it turns a fast, weak motor into a slow, strong joint, what it is predicted to hold, and what it costs in play. Drawn and checked, not yet built: every number here is a prediction from the design script.

1. Pins, lobes and a wobble

A cycloidal drive is a ring of pins and a disc with one lobe fewer than there are pins. An eccentric cam on the motor shaft throws the disc 1.0 mm off centre and swings that offset round the ring. The disc cannot spin with the cam: its lobes are trapped between the pins. So it rolls round the inside of the ring, and because it has one lobe fewer, every time the wobble goes round once the disc has turned back by one lobe.

2026-11-03T22:02:14.803851 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 22 steel pins 21-lobe disc, thrown 1.0 mm 6 output pins in 7.2 mm holes
The disc as drawn in the part files, inside its ring of 22 steel dowels on a 72 mm circle. A second disc, half a turn round, sits behind it so the two wobbles cancel.
Pins and lobes
22 pins, 21 lobes
Ratio
one lobe per motor turn: 1 / 21, so 21 : 1, output turning the other way
Disc shape
radius swings 32.4 to 34.4 mm (cut 0.08 mm inside the ideal curve): lobes 2.0 mm deep
No cusps
E · N / R = 1.0 × 22 / 36 = 0.61, under 1, so every lobe is round enough to print
Output
6 dowels on a 44 mm circle through holes of pin + 2E + clearance = 7.2 mm: the holes swallow the wobble and pass only the slow turn

2. One motor turn

Follow the dark line on the disc. The motor turns the cam a full circle; the disc's centre goes round with it, a millimetre off the axis, while the disc itself creeps backwards. After one motor turn it has moved one lobe.

2026-11-03T22:02:15.014401 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ motor 0° output 0.0° motor 120° output -5.7° motor 240° output -11.4° motor 360° output -17.1°
Snapshots at 0, 120, 240 and 360 degrees of the motor. The output turns 17.1 degrees the other way per motor turn.

3. Why a printed disc can carry it

Spur gears touch on one or two teeth at a time. Here, at any moment several lobes press on several pins, so the load is shared and the contact stress on each printed lobe stays low. That is what lets PETG carry what printed spur teeth would strip. The torque the joint can hold is the motor's, multiplied by the ratio and by the drive's efficiency, which is a guess until the joint is measured.

2026-11-03T22:02:15.109780 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 1 2 3 4 5 6 7 8 N m at the joint's output motor holding torque × 21, ideal × 0.6, predicted stall 0.34 N m 7.14 N m 4.28 N m test load 1.47 N m
Predicted, not measured: 0.34 N m is the motor's holding torque at the bench current, and the 60 percent efficiency is the number the test replaces. 4.28 N m is 2.91 times the 1.47 N m test load.

4. The price: backlash

Every gap in the load path lets the output turn a little before the load is taken: between lobe and pin, between output hole and dowel, in the cam on the motor shaft, in the dowels' fit. The gaps are in series, so they add. The prediction below is written down before the test, so the dial gauge can disagree with it.

2026-11-03T22:02:15.132922 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0.0 0.2 0.4 0.6 0.8 1.0 degrees of play at the output lobe to pin 0.22° output holes 0.26° lobes bedding in 0.08° cam play 0.10° dowel float 0.10° gate 0.9° 0.76°
The print-ready estimate from the design script: drawn clearances of 0.08 and 0.12 mm are expected to print at about 0.04 and 0.05. If the printer were exact, the same drawing would give 1.24 degrees, over the gate; printed coupons decide which.
Lobe to pin
2 × 0.04 mm at the 21 mm pitch arm = 0.22°
Output holes
2 × 0.05 mm at the 22 mm pin circle = 0.26°
Bedding in
2 × 0.015 mm of PETG flattening at the pitch arm = 0.08°
Cam play
2.0° at the cam, divided by 21 = 0.10° (an upper bound)
Dowel float
0.103° (an estimate)
Total
0.76° against the 0.9° gate

Generated from the cycloidal arm repository's design.py. Motor torque 0.34 N m is a placeholder until the bench reading; the efficiency, the clearances and the dowel float are estimates the joint test replaces.