Stanford Seminar - Robot Learning Without Action Chunking
Stanford Online · 28:51
Action chunking is the distinctive (and contested) ingredient in recent robot policies trained on human demos: predicting and executing a long action sequence from short visual context models multi-step latent style,...