Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Learning to Follow Directions in Street View
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use hard attention to select one instruction at a time | StreetNav agent · List + Goal Reward, NYC test | goal success (%): 11.3 -> 17.3 | Helped |
| Use hard attention to select one instruction at a time | StreetNav agent · List + Goal Reward, Pittsburgh test | goal success (%): 9.3 -> 12.1 | Helped |
| Remove text instructions (ablation) | StreetNav agent · List + Incremental Reward, NYC test | goal success (%): 53.6 -> 32.6 | Hurt |
| Remove thumbnail images (ablation) | StreetNav agent · List + Incremental Reward, NYC test | goal success (%): 53.6 -> 52.4 | No clear effect |
| Remove text instructions (ablation) | StreetNav agent · Step-by-Step, Pittsburgh test | goal success (%): 46.6 -> 16.9 | Hurt |
| Remove thumbnail images (ablation) | StreetNav agent · Step-by-Step, Pittsburgh test | goal success (%): 46.6 -> 38.1 | Hurt |
| Replace instruction concatenation with summation (ablation) | AllConcat · List + Incremental Reward, NYC test | goal success (%): 53.6 -> 41.6 | Hurt |
| Train first on Step-by-Step, then on List + Incremental Reward (ablation) | Hard-A · List + Incremental Reward, Pittsburgh test | goal success (%): 24.0 -> 22.5 | Mixed |
| Use ImageNet- or Places-pretrained visual features instead of learned features (ablation) | StreetNav agents · StreetNav tasks | goal success (%) | Hurt |
| Train with supervised ground-truth actions | Student and Teacher · Step-by-Step, NYC test | goal success (%): 72.7 -> 5.5 | Hurt |
Closest papers by meaning
- VLN-Video: Utilizing Driving Videos for Outdoor Vision-and-Language Navigation (AAAI 2024 · similarity 0.86)
- I Open at the Close: A Deep Reinforcement Learning Evaluation of Open Streets Initiatives (AAAI 2024 · similarity 0.86)
- CityWalker: Learning Embodied Urban Navigation from Web-Scale Videos (CVPR 2025 · similarity 0.86)
- GeoReasoner: Geo-localization with Reasoning in Street Views using a Large Vision-Language Model (ICML 2024 · similarity 0.85)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.85)
- VELMA: Verbalization Embodiment of LLM Agents for Vision and Language Navigation in Street View (AAAI 2024 · similarity 0.85)
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.