{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AZOMBBCHNG4ZQ2QSRC65OM2XFA","short_pith_number":"pith:AZOMBBCH","schema_version":"1.0","canonical_sha256":"065cc0844769b9986a1288bdd73357282eebc1b82d79cee8e1da7419ac08f39a","source":{"kind":"arxiv","id":"2503.05833","version":2},"attestation_state":"computed","paper":{"title":"Refined Policy Distillation: From VLA Generalists to RL Experts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Florian Walter, Tobias J\\\"ulg, Wolfram Burgard","submitted_at":"2025-03-06T12:52:11Z","abstract_excerpt":"Vision-Language-Action Models (VLAs) have demonstrated remarkable generalization capabilities in real-world experiments. However, their success rates are often not on par with expert policies, and they require fine-tuning when the setup changes. In this work, we introduce Refined Policy Distillation (RPD), a novel Reinforcement Learning (RL)-based policy refinement method that bridges this performance gap through a combination of on-policy RL with behavioral cloning. The core idea of RPD is to distill and refine VLAs into compact, high-performing expert policies by guiding the student policy d"},"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":"2503.05833","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2025-03-06T12:52:11Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"0cd04235e144effab37a7c30bdc28e15bc80cb29223d9604c828a6a71128a2dc","abstract_canon_sha256":"1af16f46a698e5a053c14c07537d516cb71a7e7ad32ebf8063e7e1ce58f438b0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:48:01.646301Z","signature_b64":"GcIFbU3FaDKoy4PZ87YYbT1w9cpeHeqj0z5ewiow4dPKLVtcT7ujdHOVBTAXpagwoCori/6gFu5D5Pjvtc21Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"065cc0844769b9986a1288bdd73357282eebc1b82d79cee8e1da7419ac08f39a","last_reissued_at":"2026-07-05T11:48:01.645764Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:48:01.645764Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Refined Policy Distillation: From VLA Generalists to RL Experts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Florian Walter, Tobias J\\\"ulg, Wolfram Burgard","submitted_at":"2025-03-06T12:52:11Z","abstract_excerpt":"Vision-Language-Action Models (VLAs) have demonstrated remarkable generalization capabilities in real-world experiments. However, their success rates are often not on par with expert policies, and they require fine-tuning when the setup changes. In this work, we introduce Refined Policy Distillation (RPD), a novel Reinforcement Learning (RL)-based policy refinement method that bridges this performance gap through a combination of on-policy RL with behavioral cloning. The core idea of RPD is to distill and refine VLAs into compact, high-performing expert policies by guiding the student policy d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.05833","kind":"arxiv","version":2},"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/2503.05833/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":"2503.05833","created_at":"2026-07-05T11:48:01.645825+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.05833v2","created_at":"2026-07-05T11:48:01.645825+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.05833","created_at":"2026-07-05T11:48:01.645825+00:00"},{"alias_kind":"pith_short_12","alias_value":"AZOMBBCHNG4Z","created_at":"2026-07-05T11:48:01.645825+00:00"},{"alias_kind":"pith_short_16","alias_value":"AZOMBBCHNG4ZQ2QS","created_at":"2026-07-05T11:48:01.645825+00:00"},{"alias_kind":"pith_short_8","alias_value":"AZOMBBCH","created_at":"2026-07-05T11:48:01.645825+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23640","citing_title":"Learning Process Rewards via Success Visitation Matching for Efficient RL","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18953","citing_title":"Object-Centric Residual RL for Zero-Shot Sim-to-Real VLA Enhancement","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31958","citing_title":"Adapting Generalist Robot Policies with Semantic Reinforcement Learning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17486","citing_title":"DyGRO-VLA: Cross-Task Scaling of Vision-Language-Action Models via Dynamic Grouped Residual Optimization","ref_index":96,"is_internal_anchor":false},{"citing_arxiv_id":"2502.19645","citing_title":"Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA","json":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA.json","graph_json":"https://pith.science/api/pith-number/AZOMBBCHNG4ZQ2QSRC65OM2XFA/graph.json","events_json":"https://pith.science/api/pith-number/AZOMBBCHNG4ZQ2QSRC65OM2XFA/events.json","paper":"https://pith.science/paper/AZOMBBCH"},"agent_actions":{"view_html":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA","download_json":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA.json","view_paper":"https://pith.science/paper/AZOMBBCH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.05833&json=true","fetch_graph":"https://pith.science/api/pith-number/AZOMBBCHNG4ZQ2QSRC65OM2XFA/graph.json","fetch_events":"https://pith.science/api/pith-number/AZOMBBCHNG4ZQ2QSRC65OM2XFA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA/action/storage_attestation","attest_author":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA/action/author_attestation","sign_citation":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA/action/citation_signature","submit_replication":"https://pith.science/pith/AZOMBBCHNG4ZQ2QSRC65OM2XFA/action/replication_record"}},"created_at":"2026-07-05T11:48:01.645825+00:00","updated_at":"2026-07-05T11:48:01.645825+00:00"}