<p>In 2025, VLA (Vision-Language-Action Model) became a hot topic in the field of embodied intelligence, but it has serious deficiencies in physical action execution and generalization capabilities. Two papers titled "DreamZero" and "DreamDojo," released by NVIDIA in early 2026, proposed a new paradigm that emphasizes learning from videos, enabling zero-shot generalization, breaking the limitations of data scarcity, and pointing out that the lack of a world model is the fundamental issue of VLA</p>

- In 2025, the VLA (Visual-Language-Action) model dominated the embodiment intelligence field but struggled with physical action execution and generalization beyond training environments.  
- NVIDIA introduced DreamZero and DreamDojo in early 2026, pioneering a new paradigm that allows models to learn directly from videos, demonstrating superior generalization and performance in untrained tasks.  
- This innovation shifts the focus from traditional data gathering to utilizing diverse human-generated video content, heralding a potential paradigm shift in embodied AI development.

Wallstreetcn

NVIDIA

GraniteShares 2x Short NVDA Daily ETF

Direxion Daily NVDA Bear 1X ETF

SPDR S&P Software

T-Rex 2X Long NVIDIA Daily Target ETF

AXS 1.5X NVDA Bear Daily ETF

Spdr Select Tech

VanEck Semiconductor ETF

GraniteShares 2x Long NVDA Daily ETF

YieldMax NVDA Option Income Strategy ETF

Direxion Daily NVDA Bull 2X Shares

iShares Semiconductor ETF

iShares Expanded Tech Software Sector ETF

T-Rex 2X Inverse NVIDIA Daily Target ETF

Direxion Semicon Bull 3X

XL2CSOPNVDA

XI2CSOPNVDA

Two papers from NVIDIA introduce a new paradigm of embodied intelligence after VLA