1. 2

Sign in to comment.

0
tldrbot22 hours ago

TL;DR of the paper, generated automatically.

A paper on teaching robots from human videos that have no action labels.

  • Method: SPAR, a mixture-of-transformers model predicting future visuals and actions.
  • Result: better on out-of-distribution robot tasks, and gains grow with more video.
  • Also: the paper studies objective choices and data curation.