{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-01-008",
    "source_story_id": "tmp-story-prefpi-out-of-distribution-robot-steering",
    "edition_id": "mp-2026-10-01-morning-0084",
    "edition_url": "https://themachinepress.com/edition/2026-10-01",
    "position": 8,
    "story_type": "dispatch",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "Preferences Pulled a Robot Beyond Its Original Behavior",
    "slug": "preferences-pulled-a-robot-beyond-its-original-behavior",
    "dek": "PrefPI iterated on relative judgments of self-generated trajectories instead of requiring demonstrations of the target behavior.",
    "summary": "PrefPI iterated on relative judgments of self-generated trajectories instead of requiring demonstrations of the target behavior.",
    "body_text": "PrefPI treats preferred trajectories as a conditional generative distribution and amplifies their density ratio against the broader behavior prior through classifier-free guidance. Repeating that step lets a pretrained generative policy move toward behaviors that were rare or absent initially. On real hardware, the authors report increasing object transport height from 10.7 to 19.8 centimeters using 150 preference-labeled trajectories. That measured shift demonstrates the reported task, not unconstrained out-of-distribution control.",
    "why_it_matters": "PrefPI iterated on relative judgments of self-generated trajectories instead of requiring demonstrations of the target behavior.",
    "limitations": [
      "That measured shift demonstrates the reported task, not unconstrained out-of-distribution control."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-01-008/preferences-pulled-a-robot-beyond-its-original-behavior",
    "json_url": "https://themachinepress.com/story/mp-2026-10-01-008.json",
    "first_published_at": "2026-10-01T09:00:00.000-04:00",
    "modified_at": "2026-10-01T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-01-008-001",
        "text": "PrefPI iterated on relative judgments of self-generated trajectories instead of requiring demonstrations of the target behavior.",
        "source_ids": [
          "source-2026-10-01-008"
        ],
        "qualification": "That measured shift demonstrates the reported task, not unconstrained out-of-distribution control."
      }
    ],
    "source_ids": [
      "source-2026-10-01-008"
    ],
    "tags": [
      "preference learning",
      "robot policies",
      "distribution shift"
    ],
    "image_url": "https://themachinepress.com/issues/2026-10-01/prefpi-ai-wireframe.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-01-008",
      "title": "arXiv preprint 2609.40165",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.40165",
      "canonical_url": "https://arxiv.org/abs/2609.40165",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-30T12:58:38.000-04:00",
      "accessed_at": "2026-10-01T08:18:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-01-008-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Preferences Pulled a Robot Beyond Its Original Behavior",
    "publisher": "The Machine Press",
    "published_at": "2026-10-01T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-01-008/preferences-pulled-a-robot-beyond-its-original-behavior"
  }
}