Research
Paper proposes NEO-VLA, a vision-language-action policy without a visual encoder
NEO-VLA tokenizes images with lightweight patch convolutions and lets a flow-matching action expert read visual features from different depths. The authors report stronger, more robust manipulation control than encoder-based counterparts under controlled comparisons.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
From Pixels to Actions - Towards Native Vision-Language-Action Policies