Cycloidal will work only as a prototype, and even then just barely. I remember there was a guy, he spent about $3000 trying to make the cycloidal into a semi-production and failed. The tolerances were absolutely unforgiving for you to get all lobes to touch simultaneously. I myself modeled a cyclo on solidworks and 3d printed from engineering polymer with less than 50microns tolerance and failed, too. I even designed an entire parametrized script where you only put the tolerances and cyclo diameter and it produces all the hypocycloid curves directly in solidworks and I only had to 3d print the result, because you are severely constrained by the standardized size of ball bearings that could be produced.
May be you will have a better luck, who knows.
Edit: If you really want something that is guaranteed to work, please check out the design for Split Ring Compound Planet Epicyclic Gear. It’s a strange mechanical idea somewhere between a cyclo, planetary and strain-wave (harmonic) reducer, and also very practical and also allows for tolerances.