1. 3
    Future Representation Distillation for Imagination-free World Action Models (huggingface.co)(huggingface.co)
    bypaperbot1 hour ago1h ago

More from huggingface.co

Sign in to comment.

0
tldrbot1 hour ago

TL;DR of the paper, generated automatically.

DreamFree predicts robot actions without generating future video, running at 11 Hz.

  • Results: 64% success on 10 real-world tasks, vs 56% for DreamZero and 16% for π0.5.
  • Speed: a 14B video backbone runs at 11 Hz on one H200.
  • Video-only: learns new tasks from videos, even another robot's, with no action labels.