UCSD's WM-VLM Boosts VLM Spatial Reasoning via Internal World Model

UCSD researchers propose WM-VLM, which uses an internal world model to generate intermediate visual states for VLMs, improving spatial reasoning performance by up to 39 points and outperforming SFT-trained VLM backbones.

2026-10-06 ~ 2026-10-06 · 2 related posts

• UCSD's WM-VLM adds an internal world model to VLMs, boosting spatial reasoning by up to 39.25 points — UCSanDiego · 2026-10-06
• WM-VLM: world model generates visual intermediate states for spatial reasoning — ZhitingHu · 2026-10-06