- Who it helps
- Robotics researchers comparing perception and action architectures.
- Availability
- November 4 paper is public. The laboratory repository links a 2B policy, weights, action-pretraining data and code; the training subdirectory was inaccessible in this check.
A different path from image to action
Images enter through lightweight patch convolutions, then share a backbone with language. The action component can use evolving representations at several depths instead of receiving one fixed image-encoder output.
What the comparison supports
The authors report that action-relevant information can peak before the final layer. Their controlled comparisons favor the native design, while scaling alone does not recover the gap. These are paper results, not a reproduced robot trial.
NEO-VLA lets robot actions use intermediate visual layers ↗NEO laboratory model and artifact index ↗
What you can do
- Read the linked primary evidence and its access limits before planning a trial. Open reference ↗
Where the evidence stops
Paper and repository index read November 6. Weights, dataset terms, training instructions and physical deployment were not verified.
Sources read. Sources read November 6, 2026 UTC. No third-party project, model inference or complete product workflow was executed.
Sources & further reading
Paper · paper
NEO-VLA lets robot actions use intermediate visual layers ↗
Primary page read November 6. Claims were checked against the page; the workflow was not executed.
NEO-VLA authors · page checked 2026-11-06
GitHub · implementation
NEO laboratory model and artifact index ↗
Primary page read November 6. Claims were checked against the page; the workflow was not executed.
EvolvingLMMs-Lab · page checked 2026-11-06
Coverage note: Original YouTube demonstrations and independent cross-platform attention were not established. No end-to-end workflow was executed.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
NEO-VLA studies how robot action policies can use visual information before a model’s final layer.