Before reading anything else, go and checkout this excellent project by Paul Gould. His work was very inspriational. In fact, before stumbling on this I'd never heard of a Cycloidal Drive. I'd been looking for a way to rework my stalled robot arm project as I'd been unsatisfied with the design. However, I'd not found a practical way to build a large gear reduction in a small space using planetary gears. Paul's work showed me there was another way.

I wanted to impose a few more constraints on my drive; specifically I didn't want to modify any of the purchased components and generally wanted to simplify, and reduce the costs of his design. So no custom PCBs or motor mods for me.

I designed the drive around a cheap, low kV motor, from Amazon which cost about $22. How well going cheap will work out only time will tell, but for my limited tests so far thes motors have been great. The drive would then provide a 32:1 reduction and result in a joint which is both sufficiently torque-y as well as being fast.

The drive was prototyped in PLA. I don't recommend PLA as a final material (some PETG should arrive today so I can re-print) but it's sufficiently forgiving for prototyping. And I did *a lot* of prototyping - I have a modest sized bucket of rejected parts.

There were two big problems I had to solve while designing this, and one of them was my own inability to understand what I was doing :-) Seriously, it took me ages to realize that the two cycloidal disks in the drive (you need two because each one is eccentrically mounted and requires the other to offset the eccentric momentum) needed to be offset by half-a-tooth.

The second problem was my attempts to be cheap (again). I wanted to keep the number of bearings used in the design to the absolute minimum. The two bearings to allows the eccentric motion of the cycloidal disks are unavoidable, but perhaps I could skim on the others? After all, the motor provides a bearing at one end of the input, and maybe I just needed one on the output? In practice, so long as you don't actually put a load on the drive, this worked fine. However, once any significant load was applied, the parts quickly become mis-aligned - just a tiny bit - and everything locked up. So the final design has two 6808 bearings supporting the output, two 6803 bearings supporting the input, and two 6803 bearings for the cycloidal disks. Disappointing, but works solidly under load.

The final design can be seen here (I've omitted the bearings for clarity). I'll run though the details of the design in another post.