ๆบๅจไนๅฟ JIQIZHIXIN (@jiqizhixin)
2026-01-17 | โค๏ธ 96 | ๐ 15
Why do AI models struggle to understand real-world space?
Researchers from Beihang University, Shanghai AI Lab, & Zhejiang University present a new agent, GCA, to fix it.
It forces the AI to first define a strict geometric rule from a vague question, then solve the problem only within that ruleโs bounds.
This method outperforms previous top techniques by ~27% on spatial reasoning benchmarks.
Geometrically-Constrained Agent for Spatial Reasoning
Paper: https://arxiv.org/pdf/2511.22659 Project: https://gca-spatial-reasoning.github.io/ Code: https://github.com/gca-spatial-reasoning/gca
Our report: https://mp.weixin.qq.com/s/tL7Jgd9ZYxzPhElSoPSS1A
๐ฌ PapersAccepted by Jiqizhixin
๐ ์๋ณธ ๋งํฌ
- https://arxiv.org/pdf/2511.22659
- https://gca-spatial-reasoning.github.io/
- https://github.com/gca-spatial-reasoning/gca
- https://mp.weixin.qq.com/s/tL7Jgd9ZYxzPhElSoPSS1A
๋ฏธ๋์ด
๐ Related
Auto-generated - needs manual review