Research map / Large language models / Instruction and fine-tuning
LANA: A Language-Capable Navigator for Instruction Following and Generation
Xiaohan Wang, Wenguan Wang, Jiayi Shao, Yi Yang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Jointly pretrain and fine-tune on instruction following and generation rather than using single-task agents (ablation) | LANA · R2R val unseen | SR: 65.7 -> 67.9 | Improved |
| Jointly pretrain and fine-tune on instruction following and generation rather than using single-task agents (ablation) | LANA · R2R val unseen | SPICE: 0.223 -> 0.226 | Improved |
| Add pretraining to joint-task fine-tuning (ablation) | LANA · R2R val unseen | SR: 53.1 -> 67.9 | Improved |
| Add pretraining to joint-task fine-tuning (ablation) | LANA · R2R val unseen | SPICE: 0.182 -> 0.226 | Improved |
| Add instruction generation to instruction-following training (ablation) | LANA · R2R val unseen | SR: 61.3 -> 65.7 | Improved |
| Add instruction following to instruction-generation training (ablation) | LANA · R2R val unseen | SPICE: 0.215 -> 0.223 | Improved |
| Add both temporal self-attention and visual cross-attention to the route encoder (ablation) | LANA · R2R val unseen | SR: 65.6 -> 67.9; 64.8 -> 67.9 | Improved |
| Add both temporal self-attention and visual cross-attention to the route encoder (ablation) | LANA · R2R val unseen | SPICE: 0.202 -> 0.226; 0.222 -> 0.226 | Improved |
| Use LANA mt instead of LANA st (ablation) | LANA · R2R val unseen | SPICE: 0.223 -> 0.226 | Improved |
| Use LANA mt instead of LANA st (ablation) | LANA · R4R val unseen | SPICE: 0.259 -> 0.262 | Improved |
| Train on DUET-augmented data instead of the original training data | LANA · REVERIE val unseen | SR: 34.00 -> 48.31 | Improved |
| Train on DUET-augmented data instead of the original training data | LANA · REVERIE test unseen | SR: 33.50 -> 51.72 | Mixed |
Later papers that compared with this method
1 later accepted paper reports Lana as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Controllable Navigation Instruction Generation with Chain of Thought Prompting (ECCV 2024 · better on 4 of 4 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
- Learning to Follow and Generate Instructions for Language-Capable Navigation (TPAMI 2023 · similarity 0.96)
- A New Path: Scaling Vision-and-Language Navigation With Synthetic Instructions and Imitation Learning (CVPR 2023 · similarity 0.90)
- Speaker-Follower Models for Vision-and-Language Navigation (NeurIPS 2018 · similarity 0.89)
- Bootstrapping Language-Guided Navigation Learning with Self-Refining Data Flywheel (ICLR 2025 · similarity 0.88)
- General Scene Adaptation for Vision-and-Language Navigation (ICLR 2025 · similarity 0.87)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.87)
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.
