{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:O2HCJ6MQEYIBJBYQGMHPOG7DDB","short_pith_number":"pith:O2HCJ6MQ","schema_version":"1.0","canonical_sha256":"768e24f9902610148710330ef71be31859eda9c26f23efbe0fd212d41aea8026","source":{"kind":"arxiv","id":"2507.15073","version":1},"attestation_state":"computed","paper":{"title":"Reinforcement Learning for Flow-Matching Policies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Samuel Pfrommer, Somayeh Sojoudi, Yixiao Huang","submitted_at":"2025-07-20T18:15:18Z","abstract_excerpt":"Flow-matching policies have emerged as a powerful paradigm for generalist robotics. These models are trained to imitate an action chunk, conditioned on sensor observations and textual instructions. Often, training demonstrations are generated by a suboptimal policy, such as a human operator. This work explores training flow-matching policies via reinforcement learning to surpass the original demonstration policy performance. We particularly note minimum-time control as a key application and present a simple scheme for variable-horizon flow-matching planning. We then introduce two families of a"},"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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2507.15073","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-07-20T18:15:18Z","cross_cats_sorted":[],"title_canon_sha256":"8d9535c01d71904a4c71ff3f9a64ef3f6b997d5c07c18ff973f6557820afa56e","abstract_canon_sha256":"d58a3ccb3321071d2ebb8452487cd2bf26176c7a58e206e9bf657579aae00c7e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:18.118249Z","signature_b64":"AtOmqHm+2UaZBvPIG8XyXLhWW32oefm8kegYWIBx6I9/CsqoKmC6k+GAP+AtPne7/PCARmzupEXEMNTB9FfCBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"768e24f9902610148710330ef71be31859eda9c26f23efbe0fd212d41aea8026","last_reissued_at":"2026-07-05T11:40:18.117783Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:18.117783Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reinforcement Learning for Flow-Matching Policies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Samuel Pfrommer, Somayeh Sojoudi, Yixiao Huang","submitted_at":"2025-07-20T18:15:18Z","abstract_excerpt":"Flow-matching policies have emerged as a powerful paradigm for generalist robotics. These models are trained to imitate an action chunk, conditioned on sensor observations and textual instructions. Often, training demonstrations are generated by a suboptimal policy, such as a human operator. This work explores training flow-matching policies via reinforcement learning to surpass the original demonstration policy performance. We particularly note minimum-time control as a key application and present a simple scheme for variable-horizon flow-matching planning. We then introduce two families of a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.15073","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2507.15073/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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":"2507.15073","created_at":"2026-07-05T11:40:18.117840+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.15073v1","created_at":"2026-07-05T11:40:18.117840+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.15073","created_at":"2026-07-05T11:40:18.117840+00:00"},{"alias_kind":"pith_short_12","alias_value":"O2HCJ6MQEYIB","created_at":"2026-07-05T11:40:18.117840+00:00"},{"alias_kind":"pith_short_16","alias_value":"O2HCJ6MQEYIBJBYQ","created_at":"2026-07-05T11:40:18.117840+00:00"},{"alias_kind":"pith_short_8","alias_value":"O2HCJ6MQ","created_at":"2026-07-05T11:40:18.117840+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11087","citing_title":"Test-Time Gradient Guidance of Flow Policies in Reinforcement Learning","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08602","citing_title":"Reinforcement Learning for Flow-Matching Policies with Density Transport","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17144","citing_title":"Contrastive Conceptor Activation Steering (COAST): Unlocking Vision-Language-Action Models through Hidden States","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":30,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB","json":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB.json","graph_json":"https://pith.science/api/pith-number/O2HCJ6MQEYIBJBYQGMHPOG7DDB/graph.json","events_json":"https://pith.science/api/pith-number/O2HCJ6MQEYIBJBYQGMHPOG7DDB/events.json","paper":"https://pith.science/paper/O2HCJ6MQ"},"agent_actions":{"view_html":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB","download_json":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB.json","view_paper":"https://pith.science/paper/O2HCJ6MQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.15073&json=true","fetch_graph":"https://pith.science/api/pith-number/O2HCJ6MQEYIBJBYQGMHPOG7DDB/graph.json","fetch_events":"https://pith.science/api/pith-number/O2HCJ6MQEYIBJBYQGMHPOG7DDB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB/action/storage_attestation","attest_author":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB/action/author_attestation","sign_citation":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB/action/citation_signature","submit_replication":"https://pith.science/pith/O2HCJ6MQEYIBJBYQGMHPOG7DDB/action/replication_record"}},"created_at":"2026-07-05T11:40:18.117840+00:00","updated_at":"2026-07-05T11:40:18.117840+00:00"}