The rest of day 1's talks are at: m.youtube.com/playlist?lis... and day 2 should be out soon.

These two talks really gave me something to think about as well (distributed permissions and efficient encrypted sync, with crdts):

m.youtube.com/watch?v=iLp2...
m.youtube.com/watch?v=neRu...

I think they made the right choice though: simpler is most often better, and controlling the whole stack without interop concerns lets you iterate faster and learn from your mistakes quicker.

In practice they didn't get it right first time: there is a dynamical protocol v2 that has a bigger checksum and some future-proofing reserved bits in the header etc.

When you're developing the servos, you control both sides of the communication channel, and you just want to get shit done without pulling in any 3rd party libraries.

Why dynamixel decided to go out alone? In practice, the servo bus is 3 pins daisy chained between servos: ground, 7-12V power and data (half duplex). You're not going to hang anything else off that bus, so you don't need to worry about interop.

The protocol is based on emanual.robotis.com/docs/en/dxl/... . You can think of the st3215 as a cheap dynamixel rip-off, so dynamixel compatibility is quite important.

My memory of BSG is hazy, but maybe this? m.youtube.com/watch?v=_Ctd...