{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-14-002",
    "source_story_id": "tmp-feature-datafarm-vla-alignment",
    "edition_id": "mp-2026-09-14-morning-0067",
    "edition_url": "https://themachinepress.com/edition/2026-09-14",
    "position": 2,
    "story_type": "secondary",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "The Plans Worked. The Robot Learned Almost Nothing",
    "slug": "the-plans-worked-the-robot-learned-almost-nothing",
    "dek": "Raw planner demonstrations produced 8.3% success after fine-tuning; matching the pretrained model’s motion distribution raised the result to 56.7%.",
    "summary": "Raw planner demonstrations produced 8.3% success after fine-tuning; matching the pretrained model’s motion distribution raised the result to 56.7%.",
    "body_text": "Task-and-motion planners can produce successful robot demonstrations at scale, yet DATAFARM’s authors found that feeding raw planner trajectories into a pretrained vision-language-action model transferred surprisingly little. They attribute the gap to behavior: the planner moves through joint configurations, motion styles and timing patterns unlike the data the model saw during pretraining. DATAFARM reshapes the generated demonstrations toward that prior distribution. Across three tabletop tasks and one cloth-folding task, the aligned data produced 56.7% average success, compared with 8.3% for raw planner data and 61.7% for human teleoperation. On an out-of-distribution deformable-object task, the fine-tuned model retained 85% success versus 90% before fine-tuning. These are the authors’ reported tasks and models, not a guarantee that distribution alignment replaces human robot data broadly.",
    "why_it_matters": "Raw planner demonstrations produced 8.3% success after fine-tuning; matching the pretrained model’s motion distribution raised the result to 56.7%.",
    "limitations": [
      "These are the authors’ reported tasks and models, not a guarantee that distribution alignment replaces human robot data broadly."
    ],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-14-002/the-plans-worked-the-robot-learned-almost-nothing",
    "json_url": "https://themachinepress.com/story/mp-2026-09-14-002.json",
    "first_published_at": "2026-09-14T09:00:00.000-04:00",
    "modified_at": "2026-09-14T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-14-002-001",
        "text": "Raw planner demonstrations produced 8.3% success after fine-tuning; matching the pretrained model’s motion distribution raised the result to 56.7%.",
        "source_ids": [
          "source-2026-09-14-002"
        ],
        "qualification": "These are the authors’ reported tasks and models, not a guarantee that distribution alignment replaces human robot data broadly."
      }
    ],
    "source_ids": [
      "source-2026-09-14-002"
    ],
    "tags": [
      "vision-language-action",
      "task and motion planning",
      "robot data"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-14/feature-datafarm-alignment.png",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-14-002",
      "title": "arXiv preprint 2609.12316",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.12316",
      "canonical_url": "https://arxiv.org/abs/2609.12316",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": null,
      "accessed_at": "2026-09-14T08:24:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-14-002-001"
      ]
    }
  ],
  "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 Plans Worked. The Robot Learned Almost Nothing",
    "publisher": "The Machine Press",
    "published_at": "2026-09-14T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-14-002/the-plans-worked-the-robot-learned-almost-nothing"
  }
}