Early-Signal Detector

task · first seen 4 Oct 2026

vision-language navigation

Also: VLN, VLM navigation, VLM-based navigation, language-guided navigation, vision-and-language navigation, language-conditioned navigation, vision-language model navigation

Embodied agents follow natural language instructions to navigate through visual environments.

Vision-and-language navigation is the embodied AI task in which an agent follows natural-language instructions through visual environments, typically benchmarked on Room-to-Room and its continuous-environment variants since 2018. The sources show ordinary steady output rather than a new idea. Recent papers cover aerial and UAV navigation with compact memory, parallel action decoding, episodic-memory replay for online adaptation, progress verification for long-horizon tasks, sim-to-real transfer on an Ackermann-steered robot, and an accessibility benchmark. Repositories from OpenGVLab, OpenDriveLab and others build VLN agents on native multimodal LLMs or on internet-video pretraining. The common thread is replacing purpose-built VLN models with large vision-language model backbones.

The task is a long-standing benchmark area, and the recent work mainly tests whether general multimodal models can be made efficient and reliable enough for onboard robot and drone deployment. Progress there would carry over to practical instruction-following robots.

Analyst note (established), 18 Nov 2026.

Mentions per week

01019 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 71 on 15 Nov 2026 · 22 mentions in 28 days · 22 independent groups · papers → code · first seen 6 weeks ago · acceleration +0.24

Earliest sources

Latest sources

Mentioned together with

aerial roboticssim-to-real transfervision-language-action modelzero-shot navigationrecursive self-improvementprivileged information learningagentic planning with verificationtopological graph mappingcross-modal attentioninstruction translation for navigation agents