SpatialScout tests VLM agents on active 3D scene mapping
Original title: SpatialScout: Can VLM Agents Actively Explore and Map 3D Scenes?
Claude Opus 5.5 scores 41.71% mean overlap; only 3.25% of targets exceed IoU 0.9.
SpatialScout evaluates vision-language agents that must explore indoor scenes and estimate object centers in a shared frame from RGB, calibration, poses, and an object inventory. The benchmark has 80 scenes and 3,140 target objects, scored by object-scale overlap that penalizes missing predictions. Frontier VLMs show a gap between coarse layouts and precise metric maps. Code is available on GitHub.
- The set contains 80 indoor scenes and 3,140 target objects.
- Agents work under bounded interaction budgets with viewpoint and map revision.
- Claude Opus 5.5 is the strongest model at 41.71% mean overlap.
- 47.23% of targets exceed IoU 0.5; only 3.25% exceed 0.9.
- Products
- SpatialScout
- AI models
- Claude Opus 5.5
From the original article
1 min read at huggingface.co
Spatial intelligence requires agents to actively acquire visual evidence and integrate observations into a coherent representation of the three-dimensional world. Yet spatial question answering offers limited evidence of whether models can construct accurate metric maps through autonomous exploration.
By Zhaoxi Zhang, Hao Liang, Linxuan Wu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.