I'm working on robotic wire handling. It sounds narrow until you try to automate it: wires bend, twist, slide along fixtures, and still need to end up in the right place. I'll share what I learn about perception, manipulation, and making it work beyond a demo.

DeformX caught my eye: Cosserat rod physics + Isaac Sim for wires and cables, with synthetic perception data from the same simulation. For robotic wire handling, I'd love to see how it transfers to routing through clips and around fixtures. deformx.github.io/

T-Rex has a result I keep thinking about: on its 12 tasks, naïvely adding touch to π0.5 reduced average success from 17% to 6%. The point isn’t “more sensors.” It’s learning to react to contact at a different time scale from vision. That feels central for wires and insertion.

Anthropic pays $750,000+ a year for engineers who know how to build LLMs from scratch. Stanford just released the exact lecture that teaches it - 1 hour 44 minutes, free, straight from CS229. Bookmark and watch it this weekend. It'll teach you more about how ChatGPT & Claude actually work than most people at top AI companies learn in their entire careers.