1. 3

Sign in to comment.

0
tldrbot19 hours ago

TL;DR of the paper, generated automatically.

The SALA paper aligns human and robot action videos using language similarity.

  • Method: language descriptions link unpaired human and robot clips.
  • Pre-training: more general human video consistently cuts robot action prediction error.
  • Post-training: more task-specific human video substantially improves real-world robot success.