{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-05-003",
    "source_story_id": "tmp-story-neo-vla-native-vision",
    "edition_id": "mp-2026-11-05-morning-0118",
    "edition_url": "https://themachinepress.com/edition/2026-11-05",
    "position": 3,
    "story_type": "dispatch",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "The Action Signal Leaves Before the Vision Encoder Ends",
    "slug": "the-action-signal-leaves-before-the-vision-encoder-ends",
    "dek": "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.",
    "body_text": "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.",
    "limitations": [
      "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."
    ],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-05-003/the-action-signal-leaves-before-the-vision-encoder-ends",
    "json_url": "https://themachinepress.com/story/mp-2026-11-05-003.json",
    "first_published_at": "2026-11-05T09:00:00.000-05:00",
    "modified_at": "2026-11-05T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-05-003-001",
        "text": "NEO-VLA lets robot control read evolving visual computation at multiple depths instead of inheriting one fixed encoder output.",
        "source_ids": [
          "source-2026-11-05-003"
        ],
        "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."
      }
    ],
    "source_ids": [
      "source-2026-11-05-003"
    ],
    "tags": [
      "vision-language-action",
      "robot control",
      "visual encoders"
    ],
    "image_url": "https://themachinepress.com/issues/2026-11-05/neo-vla-native-vision.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-05-003",
      "title": "arXiv preprint 2611.06918, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2611.06918v1",
      "canonical_url": "https://arxiv.org/abs/2611.06918v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-04T13:59:58.000-05:00",
      "accessed_at": "2026-11-05T08:29:24.413-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-05-003-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Action Signal Leaves Before the Vision Encoder Ends",
    "publisher": "The Machine Press",
    "published_at": "2026-11-05T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-05-003/the-action-signal-leaves-before-the-vision-encoder-ends"
  }
}