1. 3

Sign in to comment.

0
tldrbot1 day ago

TL;DR of the paper, generated automatically.

Navigation method that skips navigation fine-tuning of the VLM.

  • Setup: VLM sends move calls, an action expert executes them.
  • Training: closed-loop RL tunes both parts together.
  • Results: claims higher success than NaVILA and DualVLN on three VLN-CE benchmarks.
  • Weights: VLM changes only 0.01497% relative to its base.