{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7PFSY7BACAQBGO2XYCEXPI5XBV","short_pith_number":"pith:7PFSY7BA","schema_version":"1.0","canonical_sha256":"fbcb2c7c201020133b57c08977a3b70d7097c9773427b1da4bdbbe65063c4861","source":{"kind":"arxiv","id":"2605.17017","version":1},"attestation_state":"computed","paper":{"title":"When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Reformulating BFM task inference as a minimax problem over dynamics perturbations yields robust policies from single-environment offline data alone.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ashutosh Nayyar, Rahul Jain, Rishabh Agrawal","submitted_at":"2026-05-16T14:33:34Z","abstract_excerpt":"Behavior Foundation Models (BFMs) enable scalable imitation learning (IL) by pretraining task-agnostic representations that can be rapidly adapted to new tasks. However, existing BFMs assume fixed environment dynamics, limiting their robustness under real-world shifts such as changes in friction, actuation, or sensor noise. We address this by formulating BFM task-inference as a robust minimax optimization problem, enabling adaptation to worst-case dynamics perturbations without modifying pretraining. To the best of our knowledge, this is the first BFM-based framework that achieves robustness t"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.17017","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-16T14:33:34Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"51c09b3f440dc5b2f773b13f533f33af1896f6aa502fbce271410e6b3e50dcec","abstract_canon_sha256":"b34725124ea9eb3f03a4e0656afbb4469ad79ba6344c2cef591e9fcc1909e0c5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:36.199949Z","signature_b64":"Kiq09Jer6dBztIVIxOVGCXFPWh9aQTPwnKbmCu9PlC1/8UN4DtGqiV35Bgc3560j2pTDeVFYhfig4qEnzm5GDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fbcb2c7c201020133b57c08977a3b70d7097c9773427b1da4bdbbe65063c4861","last_reissued_at":"2026-05-20T00:03:36.199052Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:36.199052Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Reformulating BFM task inference as a minimax problem over dynamics perturbations yields robust policies from single-environment offline data alone.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ashutosh Nayyar, Rahul Jain, Rishabh Agrawal","submitted_at":"2026-05-16T14:33:34Z","abstract_excerpt":"Behavior Foundation Models (BFMs) enable scalable imitation learning (IL) by pretraining task-agnostic representations that can be rapidly adapted to new tasks. However, existing BFMs assume fixed environment dynamics, limiting their robustness under real-world shifts such as changes in friction, actuation, or sensor noise. We address this by formulating BFM task-inference as a robust minimax optimization problem, enabling adaptation to worst-case dynamics perturbations without modifying pretraining. To the best of our knowledge, this is the first BFM-based framework that achieves robustness t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"To the best of our knowledge, this is the first BFM-based framework that achieves robustness to dynamics shifts while relying solely on offline data from a single nominal environment.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The minimax optimization over dynamics perturbations can be solved tractably from offline nominal data alone and produces policies that generalize to actual (not just modeled) dynamics shifts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Robust minimax task inference in BFMs achieves dynamics-shift robustness from nominal offline data alone and outperforms standard baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Reformulating BFM task inference as a minimax problem over dynamics perturbations yields robust policies from single-environment offline data alone.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ba56163def9e78615d3ed0bb0a04cfeef81cf74c3cf7576421e95ab7c244190c"},"source":{"id":"2605.17017","kind":"arxiv","version":1},"verdict":{"id":"494bc298-6e33-492d-968e-7812db429314","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:34:04.533249Z","strongest_claim":"To the best of our knowledge, this is the first BFM-based framework that achieves robustness to dynamics shifts while relying solely on offline data from a single nominal environment.","one_line_summary":"Robust minimax task inference in BFMs achieves dynamics-shift robustness from nominal offline data alone and outperforms standard baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The minimax optimization over dynamics perturbations can be solved tractably from offline nominal data alone and produces policies that generalize to actual (not just modeled) dynamics shifts.","pith_extraction_headline":"Reformulating BFM task inference as a minimax problem over dynamics perturbations yields robust policies from single-environment offline data alone."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.17017/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.019876Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:40:42.699731Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T19:49:48.650652Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T18:51:58.944220Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.186275Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:24.869901Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"d8b42bdefe477aeb9f3c18fc52895bd94356353837d925d4dec3ed0578b59b04"},"references":{"count":68,"sample":[{"doi":"","year":2023,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","ref_index":1,"cited_arxiv_id":"2303.08774","is_internal_anchor":true},{"doi":"","year":2024,"title":"Policy optimization for strictly batch imitation learning","work_id":"6fa0b10c-a2b1-4424-b7fa-c1a60e8ea7e6","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Balance equation-based distributionally robust offline imitation learning.arXiv preprint arXiv:2511.07942, 2025","work_id":"8eecf44f-11c4-420f-a82e-26497c89989d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Markov balance satisfac- tion improves performance in strictly batch offline imitation learning","work_id":"17015320-fc1b-450b-a03e-cd47cdfa93dd","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Conditional kernel imi- tation learning for continuous state environments","work_id":"1fd3a3df-eca8-4ac0-bb01-37dc1c7fc700","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":68,"snapshot_sha256":"88b30ea67bff5af6a851c343c941134216b4888e0ab82a66f520792f667fe063","internal_anchors":6},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.17017","created_at":"2026-05-20T00:03:36.199169+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.17017v1","created_at":"2026-05-20T00:03:36.199169+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.17017","created_at":"2026-05-20T00:03:36.199169+00:00"},{"alias_kind":"pith_short_12","alias_value":"7PFSY7BACAQB","created_at":"2026-05-20T00:03:36.199169+00:00"},{"alias_kind":"pith_short_16","alias_value":"7PFSY7BACAQBGO2X","created_at":"2026-05-20T00:03:36.199169+00:00"},{"alias_kind":"pith_short_8","alias_value":"7PFSY7BA","created_at":"2026-05-20T00:03:36.199169+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV","json":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV.json","graph_json":"https://pith.science/api/pith-number/7PFSY7BACAQBGO2XYCEXPI5XBV/graph.json","events_json":"https://pith.science/api/pith-number/7PFSY7BACAQBGO2XYCEXPI5XBV/events.json","paper":"https://pith.science/paper/7PFSY7BA"},"agent_actions":{"view_html":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV","download_json":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV.json","view_paper":"https://pith.science/paper/7PFSY7BA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.17017&json=true","fetch_graph":"https://pith.science/api/pith-number/7PFSY7BACAQBGO2XYCEXPI5XBV/graph.json","fetch_events":"https://pith.science/api/pith-number/7PFSY7BACAQBGO2XYCEXPI5XBV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV/action/storage_attestation","attest_author":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV/action/author_attestation","sign_citation":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV/action/citation_signature","submit_replication":"https://pith.science/pith/7PFSY7BACAQBGO2XYCEXPI5XBV/action/replication_record"}},"created_at":"2026-05-20T00:03:36.199169+00:00","updated_at":"2026-05-20T00:03:36.199169+00:00"}