Research map / Multi-agent systems and games / Human-AI and goal-conditioned
What You See Is What You Reach: Towards Spatial Navigation with High-Level Human Instructions
Lingfeng Zhang, Haoxiang Fu, Xiaoshuai Hao, Shuyi Zhang, Qiang Zhang, Rui Liu, Long Chen, Wenbo Ding
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 |
|---|---|---|---|
| use GPT-4o instead of Qwen2.5-VL-72B for instruction reasoning (ablation) | SpNav · Spatial navigation benchmark | success rate: 28.9% -> 41.1% | Improved |
| add a VLM reasoning stage for goal extraction (ablation) | SpNav · Spatial navigation benchmark | success rate: 12.4% -> 41.1% | Improved |
| use larger open-source reasoning VLM (72B vs 7B) (ablation) | SpNav · Spatial navigation benchmark | success rate: 23.4% -> 28.9% | Improved |
| replace NaviPoint with GPT-4o for goal pointing (ablation) | SpNav · Spatial navigation benchmark | success rate: 41.1% -> 25.4% | Hurt |
| use task-specific NaviPoint (7B) instead of RoboPoint (13B) (ablation) | SpNav · Spatial navigation benchmark | success rate: 35.2% -> 41.1% | Improved |
| fine-tune pointing VLM on 200K spatial QA pairs vs general VLMs (ablation) | SpNav · Spatial navigation benchmark | success rate: 16.7% -> 41.1% (vs Qwen2.5-VL-72B) | Improved |
| use Map-to-Action instead of PointNav (ablation) | SpNav · Spatial navigation benchmark | success rate: 28.0% -> 41.1% | Improved |
| use Map-to-Action instead of Direct-and-Avoid (ablation) | SpNav · Spatial navigation benchmark | success rate: 15.0% -> 41.1% | Improved |
| use hierarchical SpNav instead of general VLMs end-to-end | SpNav vs Qwen2.5-VL-7B · Spatial navigation benchmark | success rate: 0.0% -> 41.1% | Improved |
| use SpNav instead of MapNav (best navigation-specific baseline) | SpNav · Spatial navigation benchmark | SPL: 32.1 -> 34.8 | Improved |
Closest papers by meaning
- NaVLA$^2$: A Vision-Language-Audio-Action Model for Multimodal Instruction Navigation (AAAI 2026 · similarity 0.89)
- UrbanNav: Learning Language-Guided Embodied Urban Navigation from Web-Scale Human Trajectories (AAAI 2026 · similarity 0.89)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.88)
- Scene Map-based Prompt Tuning for Navigation Instruction Generation (CVPR 2025 · similarity 0.87)
- CitySeeker: How Do VLMs Explore Embodied Urban Navigation with Implicit Human Needs? (ICLR 2026 · similarity 0.86)
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation (NeurIPS 2025 · similarity 0.86)
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.
