# MOON: Learning When to Forget and How to Recover for Vision-Language Navigation Policies

- Source: arXiv cs.RO
- Published: 2026-10-27T04:00:00.000Z
- RoboHot: https://robohot.news/en/items/lfit8x6rc36uj44w10t7gcgh3
- Original: https://arxiv.org/abs/2610.33546

## Summary

MOON is a framework that adds two lightweight modules to a frozen vision-language navigation (VLN) policy, with fewer than 1% additional parameters relative to the original policy, raising the navigation success rate of the frozen base by 5.2 and 6.0 percentage points on R2R-CE and RxR-CE, respectively.