Hi Simon, great project. I think you can improve torque by increasing the number of engaged teeth. Under no load, the teeth in the belt are engaged, but as soon as torque is applied, the belt will tend to disengage pretty quickly. 

This should be easy to fix using an elliptic wave generator, and adding some more bearings to the elliptic shape so the belt is effectively pushed to the geared walls, like the industrial harmonic drives.  

The other other limiting factor for using in robotic arms would be radial torques. For that I think some roller bearings need to be put in the design, in the outer areas, probably. 

I would love to see some torque test of this design, I'm thinking something rough, just adding a wooden stick to the output and putting some weight at the end to measure rough Nm that it can hold. 

Again, great work, I believe it has great potential and would love to contribute to it. 

Cheers!

David

At 2:18 and 2:34 you say that one side has 60 and the other has 62 teeth. The flex splne belt has 60 teeth, circular spline is 62 teeth. So the 60 tooth belt is deformed 60/62= ~3% each tooth?

After some thinking, the belt would be pushed perfectly into each cog of the the gear. The difference in the number of gears is what results in rotation. If there was a resistance then each tooth of the belt would have about 1/30 the load. The stationary 60 tooth gear/frame supplies the holding force, the 62 tooth the rotation. So the 50/50 split seems best, and the even bearing forced flex of the entire belt width is clever.