GBX-173, $160 low-resource swerve module

I’d like to preface this post by stating that if you or your team is considering doing swerve, please save up money and get some nice SDS or WCP modules. Seriously. It’s not worth the money or time to do something cheap.

With that said…

GBX-173, the Aliexpress Swerve

GBX-173, or Aliswerve, has a BOM cost of under $160 not including motors and motor controllers, with a total cost of only $1350 for four modules including motors and motor controllers. Featuring a brushless drive motor, a 1:1 absolute anti-backlash encoder, a 5.3lb assembled weight, and no machining beyond what can be achieved with a hand drill and bandsaw, this design cuts no corners on features other than reliability and strength. And really, who needs those with a price tag this low?

This design was inspired by the recent influx of “cheap” swerves to Chief Delphi. Many of these designs feature expensive gear setups, premium printer filaments, or assume access to CNC equipment. Aliswerve makes none of these assumptions, and was designed ground-up to be as cheap as possible and manufacturable with the bare minimum of tools.

Aliswerve is designed to be printed on a stock Ender 3 in PETG and contains exactly 1 “machined” part, which can be made using an included 3D-printable jig. STL files designed for printing are included in the repository. There is also a 3mm carbon steel plate that can be lasercut for $25 from Send Cut Send’s online lasercutting service.

Grabcad link here and OnShape model here. BOM is available here. The BOM is not exhaustive, but certain savings like reusing the toughbox mini 14T pinion are not assumed.

The “printable” versions are versions designed to be printed in PETG on an Ender 3 Pro with a 0.6mm nozzle, 40% infill, 3 wall layers, and 0.3mm layer height, as determined by the shrinkage measured by a friend of mine who owns an Ender 3 Pro. At these settings, one module’s worth of parts takes about 25 hours to print. If anyone has a fully stock Ender 3, I’d be interested in hearing about your experiences with repeatability and shrinkage so I can update my clearances. Holes and bosses get about a 0.25mm adder right now, depending on the size.

Design Details

At the heart of the design lies a 3mm steel baseplate.

When nested as shown, the price for 8 plates drops below $26 each. The lack of lightening ensures that you can count each piece as a 30-minute part for re-usability purposes.

The one machined part is the plate used to link the TTB bevel to the Higrip wheel.

This is a fairly difficult part to make by hand due to the six 6-32 tapped holes (Ryan pls), but I’ve included a 3D printable jig model to aid in layout and cutting. Frankly, I don’t trust this to hold up, and drilling out the bevel gear for an 8-32 screw is probably the right way to go. The outside profile doesn’t need to be very round; it can easily be a square or octagon. The bore is a simple step drilled 7/8" hole.

M4 screws, locknuts, and threaded inserts are used throughout the design due to their impressively low cost and ease of use. 6701 bearings are used where possible due to their low cost and small size.

The use of a generic all-metal 500-size planetary gearbox saves money over the more popular Versaplanetary or Ultraplanetary gearboxes favored in custom swerves.

Using metric gears, an 8mm shaft, and the WCP bevel gear set could save an extra few dollars on gears and bearings (and skips tapping 6-32 to interact with the TTB bevel), but I felt that sticking to 20DP gears was best for this design. Chinese gear and bevel gear suppliers on Aliexpress can often be unpredictable in their stock, so sticking to local sources for all gears was the safer option. I was also on the fence about using a 5mm HTD belt, but for the same reason I decided it was safer to stick to an American supplier.

A cross-section view for those curious. Just the standard swerve stackup here.

And a closeup of the anti-backlash gears. These are “sprung” so that they maintain contact with the small encoder gear 100% of the time. This is a nice technique that I think can be used more often in off-axis encoder situations.

Final Thoughts

Overall, there’s a few improvements I’d make to the structure of the module, but I’m pretty happy with how everything turned out. Would it be a great option for an FRC season? Probably not, but for years like 2018 or 2019 with few obstacles, I think something like this could work well. I’ll leave it to someone more talented to shore up any weak points I missed in the design.

Here’s a chart comparing the price of various swerve options.

Questions and comments are welcome as always.


Make that BOM cost 1323! And you got a winner.


The anti backlash gears are very cool, we actually made a very similar design to reduce backlash for the encoder gear on our shooter arm but realized it was redundant

1 Like

Well this is an interesting question. We are looking to “step up” to swerve for the 2021/22 campaign. We are a modest resource team, not in a position to throw around multiple thousands of dollars and perhaps a year or two away from maybe being able to homebrew our own design. If you were to pick one commercially available swerve drive to experiment with, which would it be? We’ll likely buy one for starters and if we like it, take the plunge for three or four more. We are also stepping up to Neos and have a bunch of PG71’s around. Cost, durability, odds that the company that make’s em will be around in a year (!), how much you can learn from working with them ( i.e. better to assemble than to just unbox). Size and weight of course factor into desirable options unless that compromises cost/durability parameters. Potential for future home brewed improvements also a plus.

How say you?

T. Wolter
Mentor and Chief of Janitorial services

1 Like

Buy an Omio first with the money you could have spent on a swerve. Students will learn a great deal from having access to CNC equipment in their designs. You want 8 modules for a a season.
The SDS and WCP modules are best.


Echoing this because I’ve said it before and I’ll say it again, the Omio was probably the single best tool investment 2220 has made. It’s let us take our student experience and our robots to another level by fully embracing CNC and CAD as integral to a fast, iterative workflow. We’ll get back to swerve eventually, but I can’t see getting swerve before the Omio as being nearly as big of a delta for the team.


Thanks. We actually do have access to high end CNC capacity. But we are in the process of moving from community based to in school, and have had a deeper crater than most regards Covid down time and talented seniors graduating. It will be a while before we can be sure of having the infrastructure (higher ability CAD students, things worked out with our new hosts, arrangements with CNC capable sponsors). I’m just looking for which existing system can be put in front of a bunch of very green builders and not scare 'em off! TW
ps, one of our sponsors makes the most advanced CNC routers for woodworking on the planet. Their success means we have to move out of their space as they need it for next gen stuff.

1 Like

GG Anand, GOATed cheap swerve. Suppliers gotta git gud and beat your price. Also I see you plugging sendcutsend, you ain’t slick. Waiting on that RAID: Shadow Legends sponsorship next.


I want the old render back.

Can you share what gearbox this is? source and price?

Also, what is the output shaft and how did you adapt that to your steering pulley? It looks like a D shaft, maybe? Did you 3D Print a pulley that integrated with that shaft? Did you use a set screw or did you capture the pulley somehow?


The BOM is a good read.


The encoder looks high quality: It is 12 bits, for only $1.78! For that price you get one chip, four resistors, two capacitors, and seven mounting pads! And it is rated five starts by two anonymous reviewers! You just need to figure out how to mount it and calibrate it since it comes with no instructions, other than each pin’s name.
We will be buying 100!


Printed pulley captured by a bearing and the gearbox. The CAD model should do a decent job of showing how it is captured.

You should really use google before posting sarcasm like this :P. The AS5600 is an excellent magnetic encoder IC that’s available for about $2 from US sources in small quantities. Check out for more information - they’re an excellent company. The 5600 is default analog output which makes it great for interfacing with any MCU. I’ve even designed and built a few encoders based around this chip. I was debating doing a custom PCB involving this for various applications in FRC, but for this project in particular the breakout will work just fine.

Usage is pretty straightforward. Give the VCC pin 3.3V, connect GND to GND, and wire the output pin to an analog input. They really are the best cheap analog encoder chips.

EDIT: some encoder chips from China are counterfeits or extra runs. The AS5145B clone I got stuck with one time didn’t have functioning indicator pins, for example, but the core functionality wasn’t compromised. You will notice I mark the magnet as a separate cost due to the one coming with the encoder being unreliable due to polarization.



There’s a lot of good money saving ideas in here. I’m going to have to review them all carefully before releasing my next Swerve design.

Things I like:

  • Wjet / laser cut steel plates instead of routered, pocketed aluminum plates are pretty cheap if you’ve got access to a commercial provider. I love that you’ve taken the time to nest them.
  • Using the 80mm ID bearing instead of the 3.5" ID x-contact bearing saves a bit
  • That’s a nice simple adapter between COTS bevel gears that don’t have a 1.875" bolt circle and standard COTS wheels
  • The anti-backlash encoder gear is awesome. I love building flexures into 3D prints, and I’m sad I didn’t think of this before!

Things that worry me:

  • I’m still not convinced that 3DP parts can stand up to the load requirements of the forks
  • I’m also not sure about the loads on the lower bearing on the vertical shaft. The reaction forces from the cantilevered small bevel gear will be pretty high.
    (I’m playing with using the Andymark Swerve and Steer U-housing for the wheel forks, since it has all the right geometry to fit your design, is all-aluminum, and is relatively cheap even compared with 3DP.)
  • On the other hand, I know from experience that a 3D print can hold the torque of a Neo as a first stage housing, so you might be able to get rid of that top steel plate if you wanted
  • I think I’d move the encoder to the top, rather than the bottom of the unit. Other than supporting the encoder, what is most of the plastic under the bottom plate there for?
  • [edit] - also consider moving the big azimuth bearing under the steel bottom plate, instead of above? That would mean another 3DP part to hold it there, but at least most of the loads would pass through the bearing, into metal, instead of into plastic (yes, I’m still cautious about 3DP parts in load bearing applications).

Once again thanks. Well done!


Oh, one more concern:

The TTB bevel gear set is 1.25 module size. Assuming a 4" wheel is traction limited at ~ 150 lbf * 2 in = 300 in * lbf = 34 N * m (if all weight was on one wheel). That means the large bevel gear is more than 5x undersized, according to KHK’s specified limit of 5.8 N * m (which I assume are based on these formulas, but I haven’t done the math myself)

The SDS and WCP modules, which have to turn 4" wheels instead of 3" wheels on the TTB, use larger, 1.5 module bevel gears. These are only ~3x undersized at a 10.3 N*m limit, which is better (although still crazy in my mind.) You might consider altering the design to use the WCP bevel gears, for a small price hike.

(note, the above limits are all “tooth bending strength” numbers. The tooth wear life limit is an order of magnitude lower. Thankfully FRC robots don’t usually see all that much wear life anyway. Aluminum gear wear notwithstanding)

1 Like

This seems like a really strange assumption. Either way there’s only one motor on the module. So the maximum torque at the bevel pinion is 3.28 (NEO stall torque) * 24/14 = 6.56 N-m or times 3 = 19.68 N-m on the driven gear. This makes it only roughly 3.4x undersized.

1 Like

+1 for AMS encoders. I’ve used the AMS AS5047P absolute encoder and my only real complaint is that they can’t be re-zeroed over SPI. I want to say the official breakout boards at ~$17 each are still cheaper than most FRC-specific options, for example:

AS5600 Breakout:
AS5047P Breakout:


That can actually happen. You’d be surprised how weird things get once you factor in things like defenders and digging into carpet. :confused:


I used the wheel friction torque limit, instead of the motor torque limit, because you can get situations where being pushed around + gearbox inefficiencies lead to higher “backdriving” torque than the motor could deliver. But that’s good to point out, thanks.

[edit] - or what @Andrew_Schreiber said