AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

NavMorph: A Self-Evolving World Model for Vision-and-Language Navigation in Continuous Environments

Xuan Yao, Junyu Gao, Changsheng Xu

ICCV 2025 · Vision-language and multimodal › Robot action and navigation · approach: Pretraining and data augmentation for VLN · 37 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add visual reconstruction loss (ablation)NavMorph (VLN-3DFF base) · R2R-CE Val Unseen (monocular, ablation)OSR: 56.88 -> 54.27 when removedImproved
add visual reconstruction loss (ablation)NavMorph (VLN-3DFF base) · R2R-CE Val Unseen (monocular, ablation)SR: 47.91 -> 45.51 when removedImproved
add action prediction loss (ablation)NavMorph (VLN-3DFF base) · R2R-CE Val Unseen (monocular, ablation)SR: 47.91 -> 44.81 when removedImproved
add KL alignment loss between posterior and prior (ablation)NavMorph (VLN-3DFF base) · R2R-CE Val Unseen (monocular, ablation)SR: 47.91 -> 44.10 when removedImproved
add self-evolution via CEM at test time (ablation)NavMorph (VLN-3DFF base) · R2R-CE Val Unseen (monocular, ablation)SR: 45.08 -> 47.91Improved
add the world model without self-evolution (ablation)VLN-3DFF · R2R-CE Val Unseen (monocular, ablation)SR: 43.77 -> 45.08Improved
use CEM forward update instead of LSTM-based evolution (ablation)NavMorph · R2R-CE Val Unseen (monocular, ablation)SR: 43.67 -> 47.91Improved
use CEM forward update instead of LSTM-based evolution (ablation)NavMorph · R2R-CE Val Unseen (monocular, ablation)testing time: 44.56s -> 21.22sImproved
use NavMorph instead of gradient-based test-time adaptation FSTTAVLN-3DFF · R2R-CE Val Unseen (monocular, ablation)SR: 43.94 -> 47.91Improved
increase CEM memory size beyond 1000 or reduce below it (ablation)NavMorph · R2R-CE Val Unseen (monocular, ablation)SR: 47.91 at Nm=1000, degrades for smaller and largerMixed
add NavMorph world modelVLN-3DFF · R2R-CE Val Unseen (monocular)SR: 43.8 -> 47.9Improved
add NavMorph world modelHNR · R2R-CE Val Unseen (panoramic)SR: 61 -> 64Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy