Zhejiang University Introduces Spatial Memory Agent for VLM Spatial Reasoning

zju · hf · 2026-08-14

Zhejiang University introduced the Spatial Memory Agent framework to enhance the spatial reasoning of Vision-Language Models (VLMs).

The approach enables a frozen VLM to self-evolve through verified experience, reflection, and reusable memory retrieval. This improvement in spatial intelligence requires no parameter updates or external tools.

Original post →

More from coding & agent

coding & agent channel →