{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-02-013",
    "source_story_id": "tmp-story-adaptive-action-chunking",
    "edition_id": "mp-2026-09-02-morning-0055",
    "edition_url": "https://themachinepress.com/edition/2026-09-02",
    "position": 13,
    "story_type": "dispatch",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "Cross-Attention Told the Robot When to Look Again",
    "slug": "cross-attention-told-the-robot-when-to-look-again",
    "dek": "A training-free rule ends an action chunk when observation grounding disperses into a sustained entropy plateau.",
    "summary": "A training-free rule ends an action chunk when observation grounding disperses into a sustained entropy plateau.",
    "body_text": "Long open-loop chunks can drift as the world changes, while short chunks repeatedly pay inference cost. The method watches cross-attention weights already computed inside the action expert and truncates execution when their entropy remains high. Evaluations across two VLA families, simulation suites and three real manipulation tasks improved average success over fixed and adaptive baselines with negligible added overhead.",
    "why_it_matters": "A training-free rule ends an action chunk when observation grounding disperses into a sustained entropy plateau.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-02-013/cross-attention-told-the-robot-when-to-look-again",
    "json_url": "https://themachinepress.com/story/mp-2026-09-02-013.json",
    "first_published_at": "2026-09-02T09:00:00.000-04:00",
    "modified_at": "2026-09-02T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-02-013-001",
        "text": "A training-free rule ends an action chunk when observation grounding disperses into a sustained entropy plateau.",
        "source_ids": [
          "source-2026-09-02-013"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-09-02-013"
    ],
    "tags": [
      "action chunking",
      "cross-attention",
      "robot control"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-02/action-chunking-code-file.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-02-013",
      "title": "arXiv preprint 2609.00908",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.00908",
      "canonical_url": "https://arxiv.org/abs/2609.00908",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": null,
      "accessed_at": "2026-09-02T08:27:36.598-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-02-013-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Cross-Attention Told the Robot When to Look Again",
    "publisher": "The Machine Press",
    "published_at": "2026-09-02T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-02-013/cross-attention-told-the-robot-when-to-look-again"
  }
}