{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-04-013",
    "source_story_id": "tmp-story-nava-wam-video-action-priors",
    "edition_id": "mp-2026-10-04-morning-0087",
    "edition_url": "https://themachinepress.com/edition/2026-10-04",
    "position": 13,
    "story_type": "dispatch",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "The Action Policy Started Learning Before Actions Were Labeled",
    "slug": "the-action-policy-started-learning-before-actions-were-labeled",
    "dek": "NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations.",
    "summary": "NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations.",
    "body_text": "NAVA-WAM learns action-relevant priors from observation-only videos instead of first learning a separate latent-action model. Future-video supervision passes through joint attention to train the action branch; later, action-labeled robot demonstrations ground that branch for control. Asymmetric attention then enables action-only inference. The preprint reports stronger in-distribution and out-of-distribution results, label efficiency and real-robot generalization across its experiments, without providing one aggregate figure in the abstract. Observation-only pretraining supplements rather than eliminates the need for labeled robot data.",
    "why_it_matters": "NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations.",
    "limitations": [
      "NAVA-WAM learns action-relevant priors from observation-only videos instead of first learning a separate latent-action model.",
      "Asymmetric attention then enables action-only inference.",
      "Observation-only pretraining supplements rather than eliminates the need for labeled robot data."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-04-013/the-action-policy-started-learning-before-actions-were-labeled",
    "json_url": "https://themachinepress.com/story/mp-2026-10-04-013.json",
    "first_published_at": "2026-10-04T09:00:00.000-04:00",
    "modified_at": "2026-10-05T08:46:40.372-04:00",
    "content_status": "updated",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-04-013-001",
        "text": "NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations.",
        "source_ids": [
          "source-2026-10-04-013"
        ],
        "qualification": "NAVA-WAM learns action-relevant priors from observation-only videos instead of first learning a separate latent-action model."
      }
    ],
    "source_ids": [
      "source-2026-10-04-013"
    ],
    "tags": [
      "world action models",
      "video pretraining",
      "robot demonstrations"
    ],
    "image_url": "https://themachinepress.com/issues/2026-10-04/nava-wam-video-priors.webp",
    "corrections": [
      {
        "correction_id": "correction-2026-10-04-001",
        "revision": 2,
        "summary": "Publication note: This October 4 edition was reconstructed on October 5 after the workstation was offline. It uses original October 2 submissions from the standard Sunday-evening arXiv release, rather than claiming those papers were available for the missed 9 AM run. The date and 9 AM publication slot preserve archive chronology; October 5 developments are excluded. The September 27 Invention Desk is carried over because its replacement was generated October 5.",
        "published_at": "2026-10-05T08:46:40.372-04:00",
        "affected_story_ids": [],
        "affected_claim_ids": []
      }
    ]
  },
  "sources": [
    {
      "source_id": "source-2026-10-04-013",
      "title": "arXiv preprint 2610.03391, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.03391v1",
      "canonical_url": "https://arxiv.org/abs/2610.03391v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-02T10:41:05.000-04:00",
      "accessed_at": "2026-10-05T08:32:10.846-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-04-013-001"
      ]
    }
  ],
  "corrections": [
    {
      "correction_id": "correction-2026-10-04-001",
      "revision": 2,
      "summary": "Publication note: This October 4 edition was reconstructed on October 5 after the workstation was offline. It uses original October 2 submissions from the standard Sunday-evening arXiv release, rather than claiming those papers were available for the missed 9 AM run. The date and 9 AM publication slot preserve archive chronology; October 5 developments are excluded. The September 27 Invention Desk is carried over because its replacement was generated October 5.",
      "published_at": "2026-10-05T08:46:40.372-04:00",
      "affected_story_ids": [],
      "affected_claim_ids": []
    }
  ],
  "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 Policy Started Learning Before Actions Were Labeled",
    "publisher": "The Machine Press",
    "published_at": "2026-10-04T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-04-013/the-action-policy-started-learning-before-actions-were-labeled"
  }
}