{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-30-009",
    "source_story_id": "tmp-story-rewam-world-action-representations",
    "edition_id": "mp-2026-09-30-morning-0083",
    "edition_url": "https://themachinepress.com/edition/2026-09-30",
    "position": 9,
    "story_type": "dispatch",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "A Robot World Model Let Control Shape What It Remembered",
    "slug": "a-robot-world-model-let-control-shape-what-it-remembered",
    "dek": "ReWAM routed action loss into a compact bottleneck while its world model learned how that state evolved.",
    "summary": "ReWAM routed action loss into a compact bottleneck while its world model learned how that state evolved.",
    "body_text": "ReWAM tests the representation shared by robot control and future-observation prediction. Feature calibration and a temporal bottleneck organize pretrained DINO features into compact states, while action-grounded shaping sends only action-loss gradients into that bottleneck. Without generative video pretraining, the authors report 93.6% success on RoboTwin 2.0; on RoboDojo they report an average score of 12.29 and 8.28% success after roughly 600 hours of embodied pretraining. Those numbers come from different evaluations and should not be read as one common scale.",
    "why_it_matters": "ReWAM routed action loss into a compact bottleneck while its world model learned how that state evolved.",
    "limitations": [
      "Feature calibration and a temporal bottleneck organize pretrained DINO features into compact states, while action-grounded shaping sends only action-loss gradients into that bottleneck.",
      "Those numbers come from different evaluations and should not be read as one common scale."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-30-009/a-robot-world-model-let-control-shape-what-it-remembered",
    "json_url": "https://themachinepress.com/story/mp-2026-09-30-009.json",
    "first_published_at": "2026-09-30T09:00:00.000-04:00",
    "modified_at": "2026-09-30T09:02:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-30-009-001",
        "text": "ReWAM routed action loss into a compact bottleneck while its world model learned how that state evolved.",
        "source_ids": [
          "source-2026-09-30-009"
        ],
        "qualification": "Feature calibration and a temporal bottleneck organize pretrained DINO features into compact states, while action-grounded shaping sends only action-loss gradients into that bottleneck."
      }
    ],
    "source_ids": [
      "source-2026-09-30-009"
    ],
    "tags": [
      "world-action models",
      "robot representations",
      "policy learning"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-30-009",
      "title": "arXiv preprint 2609.38163",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.38163",
      "canonical_url": "https://arxiv.org/abs/2609.38163",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-29T13:59:22.000-04:00",
      "accessed_at": "2026-09-30T08:24:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-30-009-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "A Robot World Model Let Control Shape What It Remembered",
    "publisher": "The Machine Press",
    "published_at": "2026-09-30T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-30-009/a-robot-world-model-let-control-shape-what-it-remembered"
  }
}