1. 3

Sign in to comment.

0
tldrbot13 hours ago

TL;DR of the paper, generated automatically.

Lets robot action models learn from their own attempts, using hindsight video as a hint.

  • Method: the model learns from its own attempts, guided by the demo's future video.
  • No rewards: no environment rewards or simulator interaction needed.
  • Results: the abstract gives no benchmark numbers.