Research map / Vision-language and multimodal / Robot action and navigation
Bridging the Visual Semantic Gap in VLN via Semantically Richer Instructions
Joaquín Ossandón, Benjamín Earle, Alvaro Soto
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 |
|---|---|---|---|
| add Speaker pre-training with synthetic instructions | Regretful-Agent · R2R val unseen | success rate: 0.43 -> 0.47 | Improved |
| add objects auxiliary task (lambda=0.5, N=2) to Speaker instruction generation | Regretful-Agent pre-training · R2R val unseen | success rate: 0.47 -> 0.51 | Improved |
| add crafted-instructions auxiliary task (beta=0.3) to Speaker | Regretful-Agent pre-training · R2R val unseen | success rate: 0.47 -> 0.50 | Improved |
| add objects auxiliary task (lambda=0.5, N=2) to Speaker | Regretful-Agent pre-training · R2R val seen | success rate: 0.71 -> 0.72 | Improved |
| increase objects per word N from 2 to 3 (lambda=0.5) | Regretful-Agent pre-training · R2R val unseen | success rate: 0.51 -> 0.47 | Hurt |
| decrease objects per word N from 2 to 1 (lambda=0.5) | Regretful-Agent pre-training · R2R val unseen | success rate: 0.51 -> 0.46 | Hurt |
| pre-train directly on rule-crafted instructions instead of Speaker output | Regretful-Agent · R2R val unseen | success rate: 0.43 -> 0.46 | Mixed |
| add objects/crafted auxiliary-task pre-training instructions | HAMT · R2R val unseen | success rate (%): 52.3 -> 52.4 | Improved |
| add Speaker pre-training | HAMT · R2R val unseen | success rate (%): 48.7 -> 52.3 | Improved |
| add visual features (vs blind agent) (ablation) | Self-Monitoring · R2R val seen | success rate: 0.32 -> 0.62 | Improved |
| add visual features (vs blind agent) (ablation) | Regretful-Agent · R2R val unseen | success rate: 0.36 -> 0.43 | Improved |
| add nouns+adjectives only (remove other words) from instructions (ablation) | Regretful-Agent · R2R val unseen | success rate: 0.43 -> 0.37 | Hurt |
Closest papers by meaning
- Learning from Unlabeled 3D Environments for Vision-and-Language Navigation (ECCV 2022 · similarity 0.91)
- Frequency-Enhanced Data Augmentation for Vision-and-Language Navigation (NeurIPS 2023 · similarity 0.91)
- Scaling Data Generation in Vision-and-Language Navigation (ICCV 2023 · similarity 0.91)
- Do Visual Imaginations Improve Vision-and-Language Navigation Agents? (CVPR 2025 · similarity 0.90)
- LangMap: A Human-Verified Benchmark for Hierarchical Open-Vocabulary Goal Navigation (NeurIPS 2026 · similarity 0.89)
- ImagineNav: Prompting Vision-Language Models as Embodied Navigator through Scene Imagination (ICLR 2025 · similarity 0.89)
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.
