robotics
The Action Signal Leaves Before the Vision Encoder Ends
NEO-VLA lets robot control read evolving visual computation at multiple depths instead of inheriting one fixed encoder output.
Summary
NEO-VLA lets robot control read evolving visual computation at multiple depths instead of inheriting one fixed encoder output.
The encoder-free design tokenizes images with lightweight patch convolutions and evolves vision jointly with language through the backbone. A flow-matching action expert can draw from instruction-conditioned visual representations at different depths. Controlled comparisons favor the native architecture over encoder-based counterparts, and full fine-tuning or extra scale narrows but does not erase the gap. Layer probes place some action-critical signals before the final visual representation. These are results across the paper's manipulation settings and distribution shifts, not a universal verdict against visual encoders.
Why it matters
NEO-VLA lets robot control read evolving visual computation at multiple depths instead of inheriting one fixed encoder output.
Limits and context
- Controlled comparisons favor the native architecture over encoder-based counterparts, and full fine-tuning or extra scale narrows but does not erase the gap.
- These are results across the paper's manipulation settings and distribution shifts, not a universal verdict against visual encoders.
Key claims
NEO-VLA lets robot control read evolving visual computation at multiple depths instead of inheriting one fixed encoder output.
Qualification: Controlled comparisons favor the native architecture over encoder-based counterparts, and full fine-tuning or extra scale narrows but does not erase the gap.
Evidence: source-2026-11-05-003
Sources
- arXiv preprint 2611.06918, original v1arXiv · primary research
Corrections
No corrections have been recorded for this story.