Starting the physical AI sprint

I'm starting this site at the same time as the hands-on part of my transition into physical AI — deliberately, rather than waiting until there's a polished project to show.

The plan for the next few weeks is Pick-and-Place using SO-101: collect my own teleoperated pick-and-place demonstrations on a SO-101 arm, fine-tune and compare two different policy architectures (ACT and SmolVLA) on the same data, and evaluate both honestly against a published reference. After that, an RL Token phase extends whichever policy comes out stronger.

I'll be updating the Pick-and-Place using SO-101 project page as real results come in, rather than writing this post again once it's done. If something doesn't work the way I expect, that will show up here too — the point of documenting this in public is the honest account, not just the finished demo.