{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WP7WP5Z7KCK4MMHBFYLYWVBCJA","short_pith_number":"pith:WP7WP5Z7","schema_version":"1.0","canonical_sha256":"b3ff67f73f5095c630e12e178b54224824a50889d454194efd2465f375bc2345","source":{"kind":"arxiv","id":"2509.04063","version":1},"attestation_state":"computed","paper":{"title":"Balancing Signal and Variance: Adaptive Offline RL Post-Training for VLA Flow Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Donglin Wang, Hongchao Lu, Hongyin Zhang, Junxi Jin, Qixin Zeng, Shiyuan Zhang, Yifan Qiao","submitted_at":"2025-09-04T09:48:43Z","abstract_excerpt":"Vision-Language-Action (VLA) models based on flow matching have shown excellent performance in general-purpose robotic manipulation tasks. However, the action accuracy of these models on complex downstream tasks is unsatisfactory. One important reason is that these models rely solely on the post-training paradigm of imitation learning, which makes it difficult to have a deeper understanding of the distribution properties of data quality, which is exactly what Reinforcement Learning (RL) excels at. In this paper, we theoretically propose an offline RL post-training objective for VLA flow models"},"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":"2509.04063","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2025-09-04T09:48:43Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"01d7aaab5b57ab65b76f8442c353ef170510a8c999163b30827225c1c40d7f12","abstract_canon_sha256":"52baf3748d42679bdcf2f641598e8da4a6454083a76113b56bed138da0b15612"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:04:52.745635Z","signature_b64":"vnDuuq+bEgNbfsbb50w8kZSavvF4ZXK8v6UUmUtfj9+Rq6PnEYGKkp1bYJhQ9y13qIxBrtQAf3naHJay1YFjCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b3ff67f73f5095c630e12e178b54224824a50889d454194efd2465f375bc2345","last_reissued_at":"2026-07-05T12:04:52.745177Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:04:52.745177Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Balancing Signal and Variance: Adaptive Offline RL Post-Training for VLA Flow Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Donglin Wang, Hongchao Lu, Hongyin Zhang, Junxi Jin, Qixin Zeng, Shiyuan Zhang, Yifan Qiao","submitted_at":"2025-09-04T09:48:43Z","abstract_excerpt":"Vision-Language-Action (VLA) models based on flow matching have shown excellent performance in general-purpose robotic manipulation tasks. However, the action accuracy of these models on complex downstream tasks is unsatisfactory. One important reason is that these models rely solely on the post-training paradigm of imitation learning, which makes it difficult to have a deeper understanding of the distribution properties of data quality, which is exactly what Reinforcement Learning (RL) excels at. In this paper, we theoretically propose an offline RL post-training objective for VLA flow models"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.04063","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/2509.04063/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":"2509.04063","created_at":"2026-07-05T12:04:52.745253+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.04063v1","created_at":"2026-07-05T12:04:52.745253+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.04063","created_at":"2026-07-05T12:04:52.745253+00:00"},{"alias_kind":"pith_short_12","alias_value":"WP7WP5Z7KCK4","created_at":"2026-07-05T12:04:52.745253+00:00"},{"alias_kind":"pith_short_16","alias_value":"WP7WP5Z7KCK4MMHB","created_at":"2026-07-05T12:04:52.745253+00:00"},{"alias_kind":"pith_short_8","alias_value":"WP7WP5Z7","created_at":"2026-07-05T12:04:52.745253+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.17486","citing_title":"DyGRO-VLA: Cross-Task Scaling of Vision-Language-Action Models via Dynamic Grouped Residual Optimization","ref_index":126,"is_internal_anchor":false},{"citing_arxiv_id":"2510.12710","citing_title":"Reflection-Based Task Adaptation for Self-Improving VLA","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA","json":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA.json","graph_json":"https://pith.science/api/pith-number/WP7WP5Z7KCK4MMHBFYLYWVBCJA/graph.json","events_json":"https://pith.science/api/pith-number/WP7WP5Z7KCK4MMHBFYLYWVBCJA/events.json","paper":"https://pith.science/paper/WP7WP5Z7"},"agent_actions":{"view_html":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA","download_json":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA.json","view_paper":"https://pith.science/paper/WP7WP5Z7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.04063&json=true","fetch_graph":"https://pith.science/api/pith-number/WP7WP5Z7KCK4MMHBFYLYWVBCJA/graph.json","fetch_events":"https://pith.science/api/pith-number/WP7WP5Z7KCK4MMHBFYLYWVBCJA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA/action/storage_attestation","attest_author":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA/action/author_attestation","sign_citation":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA/action/citation_signature","submit_replication":"https://pith.science/pith/WP7WP5Z7KCK4MMHBFYLYWVBCJA/action/replication_record"}},"created_at":"2026-07-05T12:04:52.745253+00:00","updated_at":"2026-07-05T12:04:52.745253+00:00"}