QDD Gearbox Actuator — Aaron Evans | Mechanical Engineering Project

01

Overview

QDD Gearbox Rev 01 mated to integrated motor housing

Rev 01 gearbox mated to the integrated motor housing

This started as a hackathon day in January 2026 with Russell Bilinski. We'd been looking at Caden Kraft's actuator work and wanted to learn more about how QDD gearboxes actually work by building one. Russell had a D6374 motor, an ODrive controller, bearings, and an encoder, so we started from there.

A QDD uses a low-ratio gearbox to get torque multiplication without the reflected inertia penalty of a high-ratio reduction. Keeping the ratio low helps preserve backdrivability and makes impedance control more practical, which is why low backlash and low friction mattered throughout this design. For this project, we went with a 5:1 planetary layout and a 3D-printed prototype on a Bambu P1S.

Collaboration: Russell Bilinski (previous Tesla battery design engineering intern) provided the motor, ODrive, encoder, and bearings, plus mentorship throughout. He built the workflow that takes Caden Kraft's pygeartrain gear profiles and generates STEP files, and is handling the H2 trainer and motor controller setup for testing. All CAD, mechanical design, prototyping, and test fixturing are my own work. Dyno testing is a joint effort.

Requirements

We set the initial requirements around backlash, backdrivability, torque, cost, and basic practicality for a student-built prototype.

Requirement	Target	Type
Backlash	≤ 0.5°	Hard
Cost	< $120 CAD	Hard
Backdrivability (friction torque)	< 1 Nm	Hard
Peak torque	≥ 16 Nm	Hard
Continuous torque	≥ 12 Nm	Hard
Efficiency	> 90%	Soft
Weight	< 2 kg	Soft

Project Status

Design→ CAD→ Calibration→ Rev 00A→ Rev 00B→ Requirement Tests

02

Trade Studies

We scored the major design choices with a weighted matrix on the original hackathon day before any CAD started. Gear type, ratio, and bearing type were all decided this way.

Gear Type

We compared 6 gear concepts across 7 criteria. Planetary came out on top because it was cheap, printable, and backdrivable, which were the three things that mattered most for this project.

Planetary

4.40

Sequential

4.05

Drive Belt

3.98

Cycloidal

3.73

Capstan

3.65

Strain Wave

3.65

Metric (Weight)	Planetary	Cycloidal	Belt	Capstan	Strain Wave	Sequential
Cheapness (0.20)	5	3	3.5	4	3.5	4.5
Transparency (0.15)	4	3.5	5	4	3	4
Torque Density (0.10)	4	4	3	3	5	2
Precision (0.15)	4	5	3.5	3	5	4
DFM/DFA (0.15)	4	3	4	3	3	4
Efficiency (0.10)	4.5	4	5	4.5	3.5	4
Durability (0.15)	5	4	4	4	3	5
Weighted Total	4.40	3.73	3.98	3.65	3.65	4.05

Ratio

5:1 scored highest at 4.83/5. Lower ratios are more backdrivable but give up torque multiplication; higher ratios add friction and reduce transparency. 5:1 was the sweet spot.

5:1

4.83

4:1

4.80

6:1

4.60

7:1

4.53

Bearings

Ball bearings won at 4.05/5. We weighted cost at 60% because crossed rollers scored better on performance but cost 5x more, which would have blown the $120 budget.

03

CATIA V6

The first concept pass happened in Onshape during the original hackathon day. I rebuilt the project in CATIA because I wanted to learn the CAD software used in a lot of industry work. About 60-80 hours of self-directed learning went into that assembly, much of it in rebuilding the model so future parameter changes would not break downstream parts.

The assembly is skeleton-driven: one master part holds the main parameters, reference planes, and axes. The rest of the parts reference that skeleton, so when I change a key dimension the related parts update with it.

Initial Onshape draft beside hand sketches

Initial Onshape draft from the hackathon day, beside the sketches it came from

First CATIA attempt vs. the skeleton-driven redesign

Rev 01 master assembly tree in CATIA

Simplified assembly hierarchy showing how the skeleton drives downstream parts

Rev 01 assembly render from CATIA

Rev 01 cross-section, carrier, bearings, and motor interface

Carrier sub-assembly tree from an earlier revision

Gear mesh close-up from CATIA

04

FDM Tolerancing

FDM prints aren't accurate enough for bearing fits out of the box. Bores print undersized by 0.1-0.4mm depending on diameter, so every bearing interface needs a calibrated offset in the CAD model. I keep each offset as a named parameter in the CATIA skeleton, which means the nominal bearing dimensions stay clean and I can tune each fit independently.

All bearing interfaces target a firm finger-press transition fit. For this prototype, that was enough for retention without pushing the fits tighter than needed.

Interface	Offset	Status
Housing bearing shaft	+0.10 mm	Validated
Carrier bearing bore	+0.10 mm	Validated
Lid bearing bore	+0.10 mm	Validated
Carrier output shaft	+0.10 mm	Applied
Planet bearing bore	+0.15 mm	Validated
Ring-to-housing bore	+0.30 mm	Validated

Calibration Coupon

Before printing the full build, I printed a test coupon with representative features (bearing bores, shaft posts, clearance holes) and measured each one with calipers. The results went straight back into the CATIA offsets.

Test coupon with representative bearing bores, shaft posts, and clearance holes

Feature	Nominal	Measured	Deviation	Result
Main bearing bore	37.10 mm	37.10 mm	0.00	Perfect fit
Bearing shaft post	25.10 mm	25.01 mm	-0.09	Good
Planet bore	14.00 mm	13.75 mm	-0.25	Offset increased
M3 clearance holes	3.40 mm	~3.00 mm	-0.40	Offset extrapolated

Smaller holes had worse deviation, which makes sense given FDM accuracy scales with feature size. The planet bore and clearance holes needed the largest corrections. I also ended up switching from heat-set inserts to self-tapping screws (M3: 2.6mm pilot, M4: 3.6mm, M5: 4.6mm), which was simpler and worked well for this prototype.

05

Prototyping

Rev 01 is the current assembled build. The main change is the integrated motor housing and the hardware for the dyno setup.

Rev 01 assembled with integrated motor housing

Rev 01 assembled with the integrated motor housing

Rev 01 cross-section render

Rev 01 front view

Rev 01 next to Rev 00B

Rev 01 parts laid out before assembly

Rev 01 assembled carrier closeup

Every change in this revision came from a test on Rev 00A: herringbone gears replaced with spur, carrier shoulders redesigned, lid bearing made floating, heat-set inserts swapped for self-tapping screws. All fits validated.

Rev 00B assembled with D6374 motor

Rev 00B cross-section — internal gear train

What Changed from 00A

Change	Source	Part
Herringbone → spur gears	T-001	All gears
Larger output shaft + bigger top bearing	T-001	Carrier top
Shorter top carrier walls, taller bottom walls	T-003, T-005	Carrier top/bottom
Floating lid bearing arrangement	T-002	Lid
Self-tapping threads (no heat inserts)	T-001	Carrier

The biggest driver for Rev 00B was T-001. The herringbone tooth profile was causing notchy mesh because small alignment errors between the two halves of each planet created interference. At this scale and with FDM tolerances, that tradeoff was not worth it, so I switched to spur gears for Rev 00B.

Rev 00B assembled next to Rev 00A carrier and lid

Rev 00A herringbone sun gear, replaced with spur in Rev 00B

Fit Results

Interface	Result	Status
All bearing bores	Good transition fits	Validated
Ring-to-housing	Consistent press fit	Validated
Planet bearings	All 3 consistent (fixed from 00A)	Validated
Lid drag	No parasitic drag when tightened	Resolved
Gear mesh	Smooth — spur gears eliminated notchiness	Resolved

All interfaces that failed in Rev 00A are now validated, and I've started testing against the requirements.

Backdrivability Demo

Hand backdriving the output shaft. Motor disconnected, no resistance.

Powered Operation

Rev 00B under power. D6374 motor through ODrive v3.6, tested stable up to 300 RPM output (1500 RPM motor side).

More videos: backlash measurement, assembly

Backlash Measurement

Dial indicator on output shaft, motor phases shorted for EM braking. 0° measurable backlash.

Assembly

Full assembly by hand. No special tools, just hex keys.

First complete build, printed on a Bambu P1S in PLA with off-the-shelf ball bearings. Assembled by hand and tested to validate fits and find issues.

Rev 00A assembled with lid

Rev 00A without lid, showing herringbone gear train

Fit Results

Interface	Result	Action
Housing main bearing	Slightly too loose	Offset adjusted
Lid-to-bearing	Good transition fit	No change
Planet bearings	Inconsistent (1 of 3 loose)	Tolerance review
M3 motor holes	0.13mm too tight	Bore enlarged
Sun gear bore	Very tight, needed pressing	10.05 → 10.075mm

I reprinted the carrier and housing with adjusted offsets and updated 4 skeleton parameters. But beyond the dimensional stuff, five things still felt wrong during hand testing: the lid added drag when tightened, the gear mesh felt notchy, the carrier shoulders were deforming, there was play between carrier halves, and bolt torque was too sensitive. I tracked each of those separately in Rev 00B.

Issues Found in Rev 00A

Test	Question	Finding	Decision
T-001	Why does the gearbox feel notchy?	Herringbone misalignment on at least one planet	Switch to spur gears
T-002	Is the lid over-constraining the carrier?	Yes, double-constrained via ring gear + bearing recess	Floating lid bearing
T-003	Does the printed shoulder deform?	Top carrier bearing shoulders printed poorly (overhang)	Carrier wall redesign
T-004	Can carrier halves clock under load?	Some flex when twisting by hand, not a concern operationally	No change needed
T-005	Does tightening bolts bind planets?	Narrow window; 1.1 Nm causes drag	Shoulder redesign

06

Testing

Fixture Evolution

Early dyno setup with the gearbox clamped directly and the structure built up from 2x4s

Rev 01 shifted the setup toward a more repeatable dyno fixture. We started by mounting the motor on its own so Russell could get the H2 trainer and motor controllers set up before coupling the gearbox back in. That fixturing is now figured out. The remaining work is running the motor and gearbox together against the remaining requirements.

Motor-only dyno setup for the next round of characterization

Rev 00B dyno coupling — early test setup on 2x4s

Requirement Verification

With Rev 00B assembled and the fit issues resolved, I started checking the gearbox against the project requirements.

Dial indicator measuring backlash on output shaft

T-012 backlash measurement — dial indicator on output shaft screw, motor phases shorted for EM braking

Test	Requirement	Result	Status
T-012 Backlash	≤ 0.5°	0°	PASS
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 Comparison	No major damage	—	Pending

07

What's Next

Rev 01 is assembled and installed in the motor housing. The remaining work is test execution. Backdrive torque is next, using a lever arm and kitchen scale to measure friction against the 1 Nm requirement. After that, the remaining checks are peak torque, continuous torque, efficiency, speed endurance, and the post-test health comparison. Geometry and fixturing are settled, so the project has shifted from build changes to finishing the requirement test sweep.