Robot learning in simulation
Machines Among Us
What we taught a walking robot and a robot arm in simulation, what broke, and why organizations that run fleets, floors or fields should be building this expertise now.
Talk · 34 slides · updated 2026-10-06
The slides
Videos, real training numbers and the failures, now including the SO-101 arm on NVIDIA GR00T N1.7: no release, the data fix, a learned release gate and a local vision checker. Arrow keys to move, N for speaker notes, G for the grid, F for fullscreen.
Open the slides → InteractiveThe field guide
Train a policy in your browser, design a reward and watch it get gamed, replay the full robot race, and read it all for your role.
Open the field guide →Also:The reel as a live, scrubbable pageThe full three-robot raceTrain your own policySplat Lab: one cat image, one 3D splat