Rev 01 · assembled

Planetary gearbox

A 5:1 printed reduction for a backdrivable actuator. I designed the gearbox; the motor, ODrive, and encoder came off the shelf. Rev 01 is assembled and working through the requirement tests.

5:1
Ratio
None
Measurable backlash
16Nm
Peak target
<$30
Spent of $120
3
Revisions
FDM
Manufacturing
  • 1 — Output carrier. Holds the planet pins, drives the output flange.
  • 2 — Planet gear. Bore offset +0.15 mm for a press fit.
  • 3 — Ring gear. Printed ring, internal teeth, mates to the housing.
  • 4 — Output bearing. 6805-2RS, 25 mm ID / 37 mm OD.
  • 5 — Planet bearings. 5 × 14 × 5 mm on each pin.
  • 6 — Motor. D6374 outrunner, off the shelf.
Cross-section of the gearbox mated to the motor
6 callouts · click or press ← →
02The partsRev 00B, before assembly

Three planets between a printed sun and a ring gear cut into the housing. Every bearing interface carries its own offset, because FDM bores come out undersized.

  • 1 — Gearbox housing. Carries the output bearing.
  • 2 — Housing lid. Bearing bore at +0.10 mm.
  • 3 — Carrier. Sets the planet spacing.
  • 4 — Ring gear. Internal teeth, +0.30 mm bore.
  • 5 — Sun gear. Driven off the motor shaft.
  • 6 — Planets. Three, on 5 mm bearings.
Printed gearbox parts laid out before assembly
6 callouts · click or press ← →
03What it doesRev 00B on the bench
Backdrives by hand
A low ratio keeps reflected inertia down, so the output turns without fighting the motor.
Runs under power
Driven through the ODrive at the 5:1 output.
No measurable backlash
Phases shorted for braking, output rocked against a dial indicator.
Assembles by hand
Every fit is a finger-press transition. No press, no heat-set inserts.
04Test results1 of 7 closed
Dial indicator mounted against the output shaft screw
None
Measurable play · requirement ≤ 0.5°

Dial indicator on the output shaft screw, motor phases shorted for braking. I could not find play with that gauge, so backlash is under what it reads, not zero. A mesh that tight can wear, so T-024 checks it again later.

T-012 BacklashPASS T-013 BackdriveRUNNING T-020 Peak torque— T-021 Continuous— T-022 Efficiency— T-023 Endurance— T-024 Health—
▶Full requirement table7 tests
TestRequirementResultStatus
T-012 Backlash≤ 0.5°None measurablePASS
T-013 Backdrivability< 1 Nm—In progress
T-020 Peak torque≥ 16 Nm—Pending
T-021 Continuous torque≥ 12 Nm—Pending
T-022 Efficiency> 90%—Pending
T-023 Speed endurance≥ 600 RPM—Pending
T-024 Health comparisonNo major damage—Pending

Cost and weight were set at $120 CAD and 2 kg. Spend is under $30 so far. Efficiency and weight are soft targets, the rest are hard.

05Three revisions00A → 00B → 01

Rev 00A ran, but the fits and the mesh both needed work. Closing 5 of those issues got Rev 00B under the backlash requirement, and Rev 01 then packaged the motor inside the housing and added the dyno hardware.

Printed carrier, ring gear and lid laid out beside the assembled gearbox on its motor
Gear set, ring and lid beside the gearbox assembled on its motor.
Rev 01 and Rev 00B gear sets side by side
Gear sets. Rev 01 in black, Rev 00B in white.
Rev 01 and Rev 00B assembled gearboxes side by side
Assembled. The Rev 01 body is deeper, because the motor sits inside it.
06Why planetary6 concepts, 7 criteria

Planetary won on cost, transparency, and durability, which is what mattered for a printed prototype.

Planetary4.40
Sequential4.05
Drive belt3.98
Cycloidal3.73
Capstan3.65
Strain wave3.65
▶Full scoring matrix6 × 7
Metric (weight)PlanetaryCycloidalBeltCapstanStrain waveSequential
Cheapness (0.20)533.543.54.5
Transparency (0.15)43.55434
Torque density (0.10)443352
Precision (0.15)453.5354
DFM/DFA (0.15)434334
Efficiency (0.10)4.5454.53.54
Durability (0.15)544435
Weighted total4.403.733.983.653.654.05
▶Ratio and bearings2 decisions

Lower ratios stay more backdrivable but give up torque, while higher ratios add friction and cut transparency. 5:1 scored 4.83, just ahead of 4:1 at 4.80, then 6:1 at 4.60 and 7:1 at 4.53.

Ball bearings won at 4.05. Cost carried 60% of the weight, because crossed rollers scored better on performance but cost five times more, which would have broken the $120 budget.

07Printing to fit6 interfaces
Printed calibration coupon measured with calipers

Test coupon carrying representative bearing bores, shaft posts, and clearance holes.

FDM bores print 0.1–0.4 mm undersized, and the error grows as features shrink. I printed a coupon, measured it, and fed the numbers back into CATIA as parameters.

+0.10bearing shaft +0.15planet bore +0.30ring-to-housing 5/6fits validated
▶Coupon measurements4 features
FeatureNominalMeasuredDeviationResult
Main bearing bore37.10 mm37.10 mm0.00On target
Bearing shaft post25.10 mm25.01 mm−0.09Good
Planet bore14.00 mm13.75 mm−0.25Offset increased
M3 clearance holes3.40 mm~3.00 mm−0.40Offset extrapolated

Smaller holes deviated more, which tracks with how FDM accuracy scales with feature size. I also switched from heat-set inserts to self-tapping screws: M3 into a 2.6 mm pilot, M4 into 3.6 mm, M5 into 4.6 mm.

▶All six offsetsCATIA parameters
InterfaceOffsetStatus
Housing bearing shaft+0.10 mmValidated
Carrier bearing bore+0.10 mmValidated
Lid bearing bore+0.10 mmValidated
Carrier output shaft+0.10 mmApplied
Planet bearing bore+0.15 mmValidated
Ring-to-housing bore+0.30 mmValidated

Keeping each offset as its own named parameter means the nominal bearing dimensions stay clean in the model and every fit tunes independently. All of them target a firm finger-press transition fit.

08On the dyno next6 tests open
Early dyno setup with the gearbox clamped to a wooden frame

The early setup. Rev 01 moved to a more repeatable fixture, which is what the remaining tests run on.

Backdrive torque is next, measured with a lever arm and a kitchen scale against the 1 Nm requirement. Then peak and continuous torque, efficiency, endurance, and a post-test health check.

Geometry and fixturing are settled, so what is left is running the tests rather than changing the design.

▶Why a low ratioBackdrivability

A low ratio multiplies torque without the reflected inertia a high-ratio gearbox adds, so the output stays backdrivable and force control stays practical. That is why backlash and friction matter more here than raw torque density.

▶CollaborationCredit

This started as a hackathon day in January 2026 with Russell Bilinski, after we looked at Caden Kraft's actuator work and wanted to understand the gearbox side ourselves.

Russell supplied the D6374 motor, ODrive controller, encoder, and bearings, and mentored the project throughout. He built the workflow that turns Caden Kraft's pygeartrain profiles into STEP files, and is handling the H2 trainer and motor controller setup for testing. The CAD, mechanical design, prototyping, and test fixturing are mine. Dyno testing is joint.

← All projects