{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "physical-intelligence-recap-policy-improvement-2025", "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections", "summary": "Combined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection.", "boundary": "OfflineRL, advantage conditioning and human-gated DAgger have prior work explicitly acknowledged. Acquired delta is their VLA-scale integration, not invention of each component. Long task runs and throughput are source-reported, not independently verified deployment reliability.", "area": "Robot reinforcement-learning systems", "topic": "Artificial intelligence", "year": 2025, "lane": "HUMAN", "result": "EMPIRICAL", "package_id": "physical-intelligence-recap-policy-improvement-2025", "target": { "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "content_hash": "sha256:6bf82fc5651222b6c92276c95079c3da65cd7269f2f89b83e78a9f1af98b4835" }, "humans": [ "Ali Amin", "Raichelle Aniceto", "Ashwin Balakrishna", "Kevin Black", "Ken Conley", "Grace Connors", "James Darpinian", "Karan Dhabalia", "Jared DiCarlo", "Danny Driess", "Michael Equi", "Adnan Esmail", "Yunhao Fang", "Chelsea Finn", "Catherine Glossop", "Thomas Godden", "Ivan Goryachev", "Lachy Groom", "Hunter Hancock", "Karol Hausman", "Gashon Hussein", "Brian Ichter", "Szymon Jakubczak", "Rowan Jen", "Tim Jones", "Ben Katz", "Liyiming Ke", "Chandra Kuchi", "Marinda Lamb", "Devin LeBlanc", "Sergey Levine", "Adrian Li-Bell", "Yao Lu", "Vishnu Mano", "Mohith Mothukuri", "Suraj Nair", "Karl Pertsch", "Allen Z. Ren", "Charvi Sharma", "Lucy Xiaoyang Shi", "Laura Smith", "Jost Tobias Springenberg", "Kyle Stachowicz", "Will Stoeckle", "Alex Swerdlow", "James Tanner", "Marcel Torne", "Quan Vuong", "Anna Walling", "Haohuan Wang", "Blake Williams", "Sukwon Yoo", "Lili Yu", "Ury Zhilinsky", "Zhiyuan Zhou" ], "authors_complete": true, "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261015-pi-recap-report", "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — original technical report", "url": "https://www.pi.website/download/pistar06.pdf" }, { "id": "cur20261015-support-physical-intelligence-recap-policy-improvement-2025", "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — institutional account", "url": "https://www.pi.website/blog/pistar06" } ], "points": "75000", "published_grade": { "id": "physical-intelligence-recap-policy-improvement-2025", "entry_id": "physical-intelligence-recap-policy-improvement-2025", "band": "B", "modifier": "+", "target": { "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "content_hash": "sha256:6bf82fc5651222b6c92276c95079c3da65cd7269f2f89b83e78a9f1af98b4835" }, "what_changed": "Combined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection.", "scope": "Combined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection. OfflineRL, advantage conditioning and human-gated DAgger have prior work explicitly acknowledged. Acquired delta is their VLA-scale integration, not invention of each component. Long task runs and throughput are source-reported, not independently verified deployment reliability.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "content_hash": "sha256:6bf82fc5651222b6c92276c95079c3da65cd7269f2f89b83e78a9f1af98b4835" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-16T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "organization", "name": "Physical Intelligence", "note": "Cited collaboration; the full byline is retained in the structured atlas and imported ledger, committed by hash in the narrative.", "roles": [ "discoverer" ], "source": { "citation": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — original technical report", "url": "https://www.pi.website/download/pistar06.pdf" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — original technical report", "url": "https://www.pi.website/download/pistar06.pdf" } }, { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-1", "kind": "source_reference", "source": { "citation": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — institutional account", "url": "https://www.pi.website/blog/pistar06" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "OfflineRL, advantage conditioning and human-gated DAgger have prior work explicitly acknowledged. Acquired delta is their VLA-scale integration, not invention of each component. Long task runs and throughput are source-reported, not independently verified deployment reliability.", "domain": "Robot reinforcement-learning systems" }, "sources": [ { "citation": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — original technical report", "url": "https://www.pi.website/download/pistar06.pdf" }, { "citation": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — institutional account", "url": "https://www.pi.website/blog/pistar06" } ], "statement": "# Physical Intelligence: advantage-conditioned learning from deployment and corrections\n\nCombined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection.\n\n## Scientific scope\n\nOfflineRL, advantage conditioning and human-gated DAgger have prior work explicitly acknowledged. Acquired delta is their VLA-scale integration, not invention of each component. Long task runs and throughput are source-reported, not independently verified deployment reliability.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nAli Amin; Raichelle Aniceto; Ashwin Balakrishna; Kevin Black; Ken Conley; Grace Connors; James Darpinian; Karan Dhabalia; Jared DiCarlo; Danny Driess; Michael Equi; Adnan Esmail; Yunhao Fang; Chelsea Finn; Catherine Glossop; Thomas Godden; Ivan Goryachev; Lachy Groom; Hunter Hancock; Karol Hausman; Gashon Hussein; Brian Ichter; Szymon Jakubczak; Rowan Jen; Tim Jones; Ben Katz; Liyiming Ke; Chandra Kuchi; Marinda Lamb; Devin LeBlanc; Sergey Levine; Adrian Li-Bell; Yao Lu; Vishnu Mano; Mohith Mothukuri; Suraj Nair; Karl Pertsch; Allen Z. Ren; Charvi Sharma; Lucy Xiaoyang Shi; Laura Smith; Jost Tobias Springenberg; Kyle Stachowicz; Will Stoeckle; Alex Swerdlow; James Tanner; Marcel Torne; Quan Vuong; Anna Walling; Haohuan Wang; Blake Williams; Sukwon Yoo; Lili Yu; Ury Zhilinsky; Zhiyuan Zhou\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Physical Intelligence: advantage-conditioned learning from deployment and corrections — original technical report](https://www.pi.website/download/pistar06.pdf)\n- [Physical Intelligence: advantage-conditioned learning from deployment and corrections — institutional account](https://www.pi.website/blog/pistar06)\n", "summary": "Combined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection.", "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections" }, "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "source_ref": { "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "content_hash": "sha256:6bf82fc5651222b6c92276c95079c3da65cd7269f2f89b83e78a9f1af98b4835" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:e5ed35efbe687165c12a0ba8a422c764480af22834bf893739b0c358c3facad9" }, "target": { "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "content_hash": "sha256:6bf82fc5651222b6c92276c95079c3da65cd7269f2f89b83e78a9f1af98b4835" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "physical-intelligence-recap-policy-improvement-2025", "entry_id": "physical-intelligence-recap-policy-improvement-2025", "revision": 1, "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections", "context": "Robot reinforcement-learning systems", "track": "Human-designed research method or resource", "band": "B", "modifier": "+", "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-16", "evidence_cutoff": "2026-10-16", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Combined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection.", "scope": "Combined demonstrations, autonomous rollouts and human interventions with value-based advantage estimates and conditioned policy extraction, iterating offline RL and real-robot deployment data collection. OfflineRL, advantage conditioning and human-gated DAgger have prior work explicitly acknowledged. Acquired delta is their VLA-scale integration, not invention of each component. Long task runs and throughput are source-reported, not independently verified deployment reliability.", "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. OfflineRL, advantage conditioning and human-gated DAgger have prior work explicitly acknowledged. Acquired delta is their VLA-scale integration, not invention of each component. Long task runs and throughput are source-reported, not independently verified deployment reliability.", "references": [ { "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — original technical report", "url": "https://www.pi.website/download/pistar06.pdf", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." }, { "title": "Physical Intelligence: advantage-conditioned learning from deployment and corrections — institutional account", "url": "https://www.pi.website/blog/pistar06", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." } ], "roles": [ { "who": "Source-listed authors and technical contributors", "credit": "SCIENCE", "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities." } ], "target": { "record_id": "urn:uuid:1dc67fc7-761b-48d9-a731-7989f5c0cdfe", "version": 1, "content_hash": "sha256:6bf82fc5651222b6c92276c95079c3da65cd7269f2f89b83e78a9f1af98b4835" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "75000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }