Video

Unknown · 0:00

Current vision-language-action (VLA) “generalist” robot policies look impressive in startup demos, but under standard supervised fine-tuning they mostly memorize training scenes and action chunks rather than generaliz...

Read the full summary on tuber