1. 3

Sign in to comment.

0
tldrbot1 hour ago

TL;DR of the paper, generated automatically.

ResWAM: a robot policy model that predicts only what changes in the scene.

  • Method: learns only the feature change, in frozen DINOv3 space.
  • Model: 2.3B total, with a 0.5B dual-branch DiT trained by flow matching.
  • Results: 81.1% LIBERO-Plus, 93.8% RoboTwin 2.0, at lowest compute reported.