Over the course of the last year I've invested several thousand dollars to prototype hypocycloid gearboxes and I've learned why they don't work. I couldn't tell from the outside, and it seemed like such a straightforward idea! Well, so does the Wankel engine.
I run http://marginallyclever.com/ so that I can afford to keep making robots all day. The hope is that this way I'll gain enough experience to do something truly awesome, like make a low cost robot arm, and then teach others how to do the same. When you support my stuff you don't just get a cool robot, you help make more cool robots possible. The big show stopper in my plans for world domi-i mean robot arms, is getting enough power to move the arms from the tiny 3D printer motors available everywhere. It would be so easy if I had the right gearbox available off-the-shelf. I don't, so I tried to design my own.
In a traditional gearbox like a sun/planet arrangement, the forces are evenly distributed - the planets are arranged around the sun. As torque on the sun increases, the force is transferred through all the planets at once to the outer gear. At worst it might break a tooth off one of the gears. The real downside to sun/planet gears (and also worm gears) is that they have backlash - while the input to a gear box is not moving there's still a bit of wiggle in the output. (At any given time there are only two teeth touching, and the distance between the teeth is the amount of wiggle.) That makes a robot very imprecise, which makes programming very challenging. I want joints and gearsboxes that are stiff - they stop exactly where I tell them to so the computer model and the real machine always match.
In a hypocycloid there is no backlash because there's always lots of teeth touching. Also, the force is uneven. Torque on either the output shaft or the input shaft causes the shaft to bend off-axis, away from the point of contact between the inner gear and the housing. As soon as this happens the interior gear starts to tilt. A tilted gear is a jamming gear, and I don't mean in a One Love kind of way.
With my machinist friend Bernie over at Coast Precision CNC and the incredible Jim Shook (who drafted all the designs) we attempted to lock down the parts, eliminate play, and experimented with different tolerances. It all came back to the same problem: the cycloid gear at the heart of the design is a problem hiding inside a solution.
There is one kind of cycloid gear, known as a harmonic drive, that works. It has a point of contact on opposite sides of the housing at the same time, eliminating the chance of bending. I was recently quoted $1650 CAD per unit for a combo RH-14D gearbox, DC motor, and encoder. $9900 for six is 10x more than I want to pay. I didn't bother asking what their closed-source controller costs.
So what's next? You keep sharing the Marginally Clever story and sending me your ideas, I keep making robots. Together we'll figure this out.