Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes
We propose Urbex, an agentic vision-language framework that instead casts grounding as active spatial search.
ProofPaper ↗
Key points
- City-scale 3D grounding is commonly formulated as candidate ranking, which requires precomputed instance proposals and becomes inefficient in large outdoor scenes.
- Urbex represents a city-scale 3D scene as an interactive multi-view environment, allowing a VLM agent to query landmarks, zoom into local regions, render oblique views, and commit a grounded bounding box.
- We optimize this tool-use policy with reinforcement learning on CityRefer, using smooth localization rewards and lightweight shaping rewards for efficient exploration.
- Zero-shot results on STPLS3D-Refer further indicate improved cross-scene robustness.
Sources (2)
- [1]Urbex: Agentic Spatial Grounding for City-Scale 3D ScenesHugging Face Daily Papers · Oct 13, 12:00 AM
We propose Urbex, an agentic vision-language framework that instead casts grounding as active spatial search.
City-scale 3D grounding is commonly formulated as candidate ranking, which requires precomputed instance proposals and becomes inefficient in large outdoor scenes.
- [2]Urbex: Agentic Spatial Grounding for City-Scale 3D ScenesarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 01:23 PM · same content
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026Neos: Learning to Interleave Search and Visual Reasoning
- Oct 12, 2026Avoiding What Your Teacher Already Knows: Entity Familiarity for Deep-Search Distillation
- Oct 12, 2026Of Errors and Echoes: Responsibility-Aware Credit Assignment for Tool-Use Agents
- Oct 11, 2026A Sovereign, Indic-First, Reasoning Model
- Oct 11, 2026RSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-Evolution
- Oct 10, 2026Self-Evolving Visual Agent via Margin Reward
