Hugging Face Papers: NEO-VLA, an encoder-free vision-language-action framework, introduced

ResearchPublished: Source: Hugging Face Papers

#Hugging Face

Key point

NEO-VLA, a framework that directly accesses visual computation without compressing information via encoders, has been introduced. It received 55 upvotes on Hugging Face Daily Papers.

Conventional Vision-Language-Action (VLA) models are typically built on the premise that perception is first compressed by visual encoders, with action reasoning beginning only afterwards. However, manipulation often depends on visual signals most vulnerable to this abstraction, such as local geometry, boundaries, depth, and subtle state changes.

In response, NEO-VLA has been introduced as a structurally encoder-free vision-language-action framework. This approach aims to allow action policies to directly access evolving visual computation rather than inheriting a fixed visual endpoint. The paper received 55 upvotes on Hugging Face Daily Papers.

This item is a summary. The full, checkable version is at the original source:

Read the original at Hugging Face Papers ↗