I reached out to Cornell, Michigan and Monash with a couple of questions regarding the fine details of prototyping cycloidal drives. They have been super useful - here are there responses:

List of Questions

How precise of a machining tolerance did you use when fabricating the gearboxes?
Were you able to machine these in house, or did you need to outsource to high precision machining services like wire EDM?
How did you generate the outer profile for the cycloidal disks? The main resource I've found online is a SolidWorks blog guide which provides a parametric equation for the disk's profile using a couple of variables as inputs, but I was wondering if your team designed it differently
What did your prototyping process look like?
Any material recommendations for the gearboxes?
Do you have any overall tips or recommended resources for designing and machining these gearboxes?

Monash Responses - From Ayden Monsant

Unfortunately I don't have an actual number anymore, and it's gonna vary depending on the CNC used. We needed to fabricate a bunch of them, so we messed around with a tool offset while leaving the cycloidal ring and discs at a nominal size. We did the ring first (our pins are fixed in place and do not rotate. We call this structure the ring) and used this as a guide when doing the discs as shown in the pic below. We really just ended up feeling the sliding fit between these two components. Start at nominal and re-do your final pass with an increasing tool offset if the ring is too tight a fit.
We managed to do these mostly wherever we could source some donated machining hours. We don't have direct access to a CNC but can ask our university machine workshop near us to machine them for us, but this was very expensive. While they have Wire EDM capabilities, these parts were done entirely with conventional CNC. We also used a few external sponsors who donated some machinist hours to helping us get these done. You could also technically machine the discs using a custom ground single-point gear cutter, but this would take a long time and a lot of patience!
Yep, that exact Solidworks tutorial as well as a couple of other papers were very useful in disc design. I have attached a paper that goes into a bit more mathematical detail of where these equations come from, as well as the impact of eccentricity and pin-size on a disc design. Our design uses a 50:1 ratio and 3mm pins with a 0.75mm eccentricity.
We started prototyping purely out of 3D printed FDM parts, initially from testing the basic operating principle and eventually to a completely plastic mock-up of the gearboxes with motors and even arm linkages. If you have a look at our arm footage in our 2020 SAR, we were running fully plastic 3d printed gearboxes on our base rotation and elbow joints, including all shafts. From here we did one more design iteration before going all-metal. It's super important that you test every small component as soon as you can within a suitable loading environment, and there's nothing better than testing within an actual gearbox. I did maybe 2 dozen iterations and printed every second one. Expect lots of iteration!
We went with mostly Aluminium 7075 for our high-wear components (cycloidal rings and discs) and some 6063 for other non-wearing structural components. We used some regular silver steel for our gear shafts, quenched and tempered to get them as hard as our bearings, and we used leaded bronze for our sliding eccentric bushings, although brass could also work. The main wearing components in a cycloidal gearbox are of course the disc contour and the pins on the ring, so make sure the two materials used here are wear resistant. 7075 is not ideal from a long-term life perspective, but we needed to keep the mass down as much as possible. So far after a lot of use, there is still no sign of wear however. Your eccentric bearing/bushing mechanism might also need wear-consideration, ours is split into three to reduce overall loading per eccentric bushing.
Consider a few different configurations of cycloidal gearboxes out there. My final design is a simplified and more lightweight version of what Nabtesco use in their cycloidal gearboxes. The centrally driven ones seem simpler, and while we used this kind of design in our smaller wrist gearboxes, I find this design a lot more difficult to modify and adapt. We've had a fair few calibration and adjustment issues with this design. I cannot stress enough that you should just spend a lot of time fully printing and testing printed gearboxes. I learned a lot from doing this, making the final metal ones much better designed. Once you get the rings and discs machined, the rest can often be done using manual lathes and mills if you are trained to do so, and you can save a lot of money doing this.

Cornell Responses - From Matthew Sherman

Precision is the name of the game in the manufacture of these gearboxes. Traditional spurs gears have a fixed radial position between gears and thus require at least a little tooth to tooth clearance to prevent binding. Because the cycloid both rotates and translates it can radially translate into mesh and as a result doesn't require any clearance. Threading the needle on "zero clearance" is a daring process though. Controlling position and orientation of gear tooth features relative to other datums on the gear (such as bearing bore) becomes extremely difficult if you try to design for virtually zero clearance. This is because the tolerance stack doesn't close for any reasonable tolerance band, in other words, there is overlap between the inner and outer gear teeth and the gearbox would just literally no longer fit together. That being said, I think it's pretty intuitive that the output error is in some way proportional to radial clearance and so a reasonable minimization is critical, and this process can be done pretty straightforwardly by just making a tolerance stack-up in Excel. This is just one simple example of critically controlled features and the need to understand all the accumulated tolerances. There are a couple features per part that can have similar "needs to be really precisely controlled, but the more we shrink the tolerance band, the more risk we run on things no longer fitting together".
We machine these in house on a Haas VF2 because outsourcing is usually super expensive, costly, and unless you convey very well what is critical and what is not through GD&T, the parts don't come back as expected. That being said, several parts could really benefit from wire EDM, especially the cycloidal gear.
The cycloidal profile can be generated parametrically in Matlab or Python. Then the plots can be exported in a format that is convertible to .dxf which your CAD package should be able to import and use in a sketch.
Hard to prototype because most of the difficulty in making these things come from the nuance manufacturing method. So like 3D printing one of these doesn't really tell you very much. Just design one and try making it, when you run into issues assembling or running, modify the manufacturing process accordingly.
Surface hardenable steels. We used nitrided 4140 for a little bit but have now moved to a tool steel which we harden in house and machine in a hard state.
There really are no "quick tips" to getting this to work. If you want some more info we can set up a zoom call because I'm too lazy to type out everything I could tell you about this. I've really hardly even scratched any surface at all with the info I've given you here. Let me know if you want to zoom to dicuss the rationale beyond different design aspects and material choices. If not, no hard feelings, again sorry it took so long to get back to you, it just totally slipped my mind.

Michigan Responses - From Vasil Iakimovitch

We generally machined components to within one thou tolerance for critical features, as this is about the tightest tolerance we can reasonably hold with our in-house machines.
We were able to make the gearboxes entirely in-house, primarily using a Haas VF-2 CNC and a manual lathe. We honestly didn't really consider outsourcing our components because of the associated cost and time.
Yep, we used that SolidWorks blog post to generate the parametric equations for the disk profiles.
We initially built and load tested a prototype gearbox that had the key features milled into a simple block of aluminum. We iterated on that design a bit before manufacturing the final gearboxes for URC 2020. Based on our experiences with the URC 2020 gearboxes, we made some design changes and machined new gearboxes for URC 2021.
We've mainly used 6061 aluminum and Delrin in the past, with some components being made from 7075 aluminum or 1018 steel to better withstand cyclic loading. If you want optimal wear resistance, you will want to stick with the harder materials, but that will of course come with the tradeoff of greater mass.
From what we've found, the performance of cycloid gearboxes is very dependent on maintaining tight machining tolerances and proper clearances between parts in the gearbox. My suggestion is to pay close attention to designing for manufacturing. You may also need to do some testing to dial in the right clearances for your gearboxes.

Browser not supported