MOON: Learning When to Forget and How to Move On in Vision-and-Language Navigation
MOON: Learning When to Forget and How to Move On in Vision-and-Language Navigation
This research presents MOON, a framework that equips frozen vision-and-language navigation policies with lightweight modules for detecting deviations and recovering from them. By clearing visual history upon deviation detection and training a recovery policy via GRPO, MOON improves success rates on R2R-CE and RxR-CE benchmarks by 5.2 and 6.0 points respectively, outperforming compared methods without retraining the base policy.
The paper introduces MOON, a lightweight framework that adds deviation detection and recovery modules to frozen vision-and-language navigation policies without retraining, achieving the highest success rates on R2R-CE and RxR-CE benchmarks while improving zero-shot object-goal navigation on HM3D-OVON.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- code: Available Artifact link Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
with GRPO from alarm states, learns how to move on, turning more of the successes found only by sampling into a single greedy rollout. Experiments on R2R-CE and RxR-CE show that MOON reaches the highest R2R-CE success rate among the compared methods and raises the success rate of its frozen base by 5.2 and 6.0 points, and an ablation with a random trigger indicates that this gain depends on when the history is cleare
Open source S4
less. After a deviation, the retained history keeps the policy from turning where a turn is needed, which we call behavioral inertia. We propose MOON, a framework that equips a frozen VLN policy with two lightweight modules, adding less than 1% of its parameters. A deviation judge decides when to forget, raising an alarm that clears the history when it predicts that the agent is off route. A recovery policy, trained
Open source S3
d. Without retraining, MOON also improves zero-shot object-goal navigation on HM3D-OVON. Together, these results suggest that deviation recovery depends on knowing when to forget, not only how to act. Code is available at https://github.com/AutoLab-SAI-SJTU/MOON.
Open source S5
arXiv:2610.33546v2 Announce Type: replace Abstract: Recent vision-and-language navigation (VLN) policies act on their visual history, and an incorrect action can take them off the instructed route. Paired interventions on four VLN policies show that clearing the history doubles pooled success at the onset of a deviation but lowers it on route, while later in a deviation, removing only the off-route frames helps much
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org