{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:QRJCWFWSL5PT5SNWIBBQZSRBZ2","short_pith_number":"pith:QRJCWFWS","canonical_record":{"source":{"id":"2602.09580","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.RO","submitted_at":"2026-02-10T09:28:20Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"39b6474a51c5644c811b32010c693ec36bef84e32b92ce77b3841cd91803a06c","abstract_canon_sha256":"8291ebe3efb7963e20c30273da9e391be94e540c40c4dd1a33556da20206d125"},"schema_version":"1.0"},"canonical_sha256":"84522b16d25f5f3ec9b640430cca21ce94c31b45e5491e8f191e309a96ec831f","source":{"kind":"arxiv","id":"2602.09580","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.09580","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"arxiv_version","alias_value":"2602.09580v4","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.09580","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"pith_short_12","alias_value":"QRJCWFWSL5PT","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"pith_short_16","alias_value":"QRJCWFWSL5PT5SNW","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"pith_short_8","alias_value":"QRJCWFWS","created_at":"2026-06-08T01:05:07Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:QRJCWFWSL5PT5SNWIBBQZSRBZ2","target":"record","payload":{"canonical_record":{"source":{"id":"2602.09580","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.RO","submitted_at":"2026-02-10T09:28:20Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"39b6474a51c5644c811b32010c693ec36bef84e32b92ce77b3841cd91803a06c","abstract_canon_sha256":"8291ebe3efb7963e20c30273da9e391be94e540c40c4dd1a33556da20206d125"},"schema_version":"1.0"},"canonical_sha256":"84522b16d25f5f3ec9b640430cca21ce94c31b45e5491e8f191e309a96ec831f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-08T01:05:07.210673Z","signature_b64":"3J70s2L6kotKfbZz2Zi5YNWu7pHvnajhSAtZxWqoNSmHz+UlvKt4hFOaaGlKqzhBsYLs0MYM8gLBoLTCeGY/CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"84522b16d25f5f3ec9b640430cca21ce94c31b45e5491e8f191e309a96ec831f","last_reissued_at":"2026-06-08T01:05:07.209647Z","signature_status":"signed_v1","first_computed_at":"2026-06-08T01:05:07.209647Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2602.09580","source_version":4,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-08T01:05:07Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"8UfsDJ/lA+fooG5wq0FobMW7qpcCHQda3Z3U209jhUJPazEBz2Ki7C1iLVMGKQu9bCmlPKNauj7367DlyuwqBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-09T05:05:31.020109Z"},"content_sha256":"7db8a21daf02b0d3d1aa050e35cda903cc269564097cfc9ce0d0ee43ef075243","schema_version":"1.0","event_id":"sha256:7db8a21daf02b0d3d1aa050e35cda903cc269564097cfc9ce0d0ee43ef075243"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:QRJCWFWSL5PT5SNWIBBQZSRBZ2","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SERNF: Sample-Efficient Real-World Dexterous Policy Fine-Tuning via Action-Chunked Critics and Normalizing Flows","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SERNF achieves sample-efficient real-world fine-tuning of multimodal dexterous policies by pairing exact-likelihood normalizing flow policies with action-chunked value critics.","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Chenyu Yang, Davide Liconti, Denis Tarasov, Hehui Zheng, Robert K. Katzschmann, Romain Guntz","submitted_at":"2026-02-10T09:28:20Z","abstract_excerpt":"Real-world fine-tuning of dexterous manipulation policies remains challenging due to limited real-world interaction budgets and highly multimodal action distributions. Diffusion-based policies, while expressive, do not permit conservative likelihood-based updates during fine-tuning because action probabilities are intractable. In contrast, conventional Gaussian policies collapse under multimodality, particularly when actions are executed in chunks, and standard per-step critics fail to align with chunked execution, leading to poor credit assignment. We present SERFN, a sample-efficient off-pol"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"To our knowledge, this is the first demonstration of a likelihood-based, multimodal generative policy combined with chunk-level value learning on real robotic hardware.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That normalizing flows can be trained to produce stable, exact likelihoods for multimodal action chunks under real-world noise and limited samples, and that the action-chunked critic will produce value estimates that align with the policy's temporal execution without introducing bias.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SERNF achieves sample-efficient real-world fine-tuning of multimodal dexterous policies by pairing exact-likelihood normalizing flow policies with action-chunked value critics.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"72a5fa406b768d104c261c09cf0ab001fdc4c5517c1876a7257389b4a9984a16"},"source":{"id":"2602.09580","kind":"arxiv","version":4},"verdict":{"id":"480e3de1-72f8-4040-aded-dbee9276b131","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T03:28:59.251908Z","strongest_claim":"To our knowledge, this is the first demonstration of a likelihood-based, multimodal generative policy combined with chunk-level value learning on real robotic hardware.","one_line_summary":"SERNF achieves sample-efficient real-world fine-tuning of multimodal dexterous policies by pairing exact-likelihood normalizing flow policies with action-chunked value critics.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That normalizing flows can be trained to produce stable, exact likelihoods for multimodal action chunks under real-world noise and limited samples, and that the action-chunked critic will produce value estimates that align with the policy's temporal execution without introducing bias.","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.09580/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"},"verdict_id":"480e3de1-72f8-4040-aded-dbee9276b131"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-08T01:05:07Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1APbklCRYlOTECyRFRIsXW/+ht3SNFDgHZJ/F59n9i+oPzZPXqsUSjFpW6nLFkh0UZC4xVBRNNjwPkb1jB1qAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-09T05:05:31.020569Z"},"content_sha256":"43bb896397b355ef3413dee322bab4eae702c5d489314a30098f4b63f6583ea3","schema_version":"1.0","event_id":"sha256:43bb896397b355ef3413dee322bab4eae702c5d489314a30098f4b63f6583ea3"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2/bundle.json","state_url":"https://pith.science/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-09T05:05:31Z","links":{"resolver":"https://pith.science/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2","bundle":"https://pith.science/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2/bundle.json","state":"https://pith.science/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2/state.json","well_known_bundle":"https://pith.science/.well-known/pith/QRJCWFWSL5PT5SNWIBBQZSRBZ2/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:QRJCWFWSL5PT5SNWIBBQZSRBZ2","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"8291ebe3efb7963e20c30273da9e391be94e540c40c4dd1a33556da20206d125","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.RO","submitted_at":"2026-02-10T09:28:20Z","title_canon_sha256":"39b6474a51c5644c811b32010c693ec36bef84e32b92ce77b3841cd91803a06c"},"schema_version":"1.0","source":{"id":"2602.09580","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.09580","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"arxiv_version","alias_value":"2602.09580v4","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.09580","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"pith_short_12","alias_value":"QRJCWFWSL5PT","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"pith_short_16","alias_value":"QRJCWFWSL5PT5SNW","created_at":"2026-06-08T01:05:07Z"},{"alias_kind":"pith_short_8","alias_value":"QRJCWFWS","created_at":"2026-06-08T01:05:07Z"}],"graph_snapshots":[{"event_id":"sha256:43bb896397b355ef3413dee322bab4eae702c5d489314a30098f4b63f6583ea3","target":"graph","created_at":"2026-06-08T01:05:07Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":3,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"To our knowledge, this is the first demonstration of a likelihood-based, multimodal generative policy combined with chunk-level value learning on real robotic hardware."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That normalizing flows can be trained to produce stable, exact likelihoods for multimodal action chunks under real-world noise and limited samples, and that the action-chunked critic will produce value estimates that align with the policy's temporal execution without introducing bias."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SERNF achieves sample-efficient real-world fine-tuning of multimodal dexterous policies by pairing exact-likelihood normalizing flow policies with action-chunked value critics."}],"snapshot_sha256":"72a5fa406b768d104c261c09cf0ab001fdc4c5517c1876a7257389b4a9984a16"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2602.09580/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Real-world fine-tuning of dexterous manipulation policies remains challenging due to limited real-world interaction budgets and highly multimodal action distributions. Diffusion-based policies, while expressive, do not permit conservative likelihood-based updates during fine-tuning because action probabilities are intractable. In contrast, conventional Gaussian policies collapse under multimodality, particularly when actions are executed in chunks, and standard per-step critics fail to align with chunked execution, leading to poor credit assignment. We present SERFN, a sample-efficient off-pol","authors_text":"Chenyu Yang, Davide Liconti, Denis Tarasov, Hehui Zheng, Robert K. Katzschmann, Romain Guntz","cross_cats":["cs.LG"],"headline":"SERNF achieves sample-efficient real-world fine-tuning of multimodal dexterous policies by pairing exact-likelihood normalizing flow policies with action-chunked value critics.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.RO","submitted_at":"2026-02-10T09:28:20Z","title":"SERNF: Sample-Efficient Real-World Dexterous Policy Fine-Tuning via Action-Chunked Critics and Normalizing Flows"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.09580","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-16T03:28:59.251908Z","id":"480e3de1-72f8-4040-aded-dbee9276b131","model_set":{"reader":"grok-4.3"},"one_line_summary":"SERNF achieves sample-efficient real-world fine-tuning of multimodal dexterous policies by pairing exact-likelihood normalizing flow policies with action-chunked value critics.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"","strongest_claim":"To our knowledge, this is the first demonstration of a likelihood-based, multimodal generative policy combined with chunk-level value learning on real robotic hardware.","weakest_assumption":"That normalizing flows can be trained to produce stable, exact likelihoods for multimodal action chunks under real-world noise and limited samples, and that the action-chunked critic will produce value estimates that align with the policy's temporal execution without introducing bias."}},"verdict_id":"480e3de1-72f8-4040-aded-dbee9276b131"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:7db8a21daf02b0d3d1aa050e35cda903cc269564097cfc9ce0d0ee43ef075243","target":"record","created_at":"2026-06-08T01:05:07Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"8291ebe3efb7963e20c30273da9e391be94e540c40c4dd1a33556da20206d125","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.RO","submitted_at":"2026-02-10T09:28:20Z","title_canon_sha256":"39b6474a51c5644c811b32010c693ec36bef84e32b92ce77b3841cd91803a06c"},"schema_version":"1.0","source":{"id":"2602.09580","kind":"arxiv","version":4}},"canonical_sha256":"84522b16d25f5f3ec9b640430cca21ce94c31b45e5491e8f191e309a96ec831f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"84522b16d25f5f3ec9b640430cca21ce94c31b45e5491e8f191e309a96ec831f","first_computed_at":"2026-06-08T01:05:07.209647Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-08T01:05:07.209647Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"3J70s2L6kotKfbZz2Zi5YNWu7pHvnajhSAtZxWqoNSmHz+UlvKt4hFOaaGlKqzhBsYLs0MYM8gLBoLTCeGY/CQ==","signature_status":"signed_v1","signed_at":"2026-06-08T01:05:07.210673Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.09580","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:7db8a21daf02b0d3d1aa050e35cda903cc269564097cfc9ce0d0ee43ef075243","sha256:43bb896397b355ef3413dee322bab4eae702c5d489314a30098f4b63f6583ea3"],"state_sha256":"c7a599d4bf3e653017bdcb5aaa21e6ad6c326204715c4522f4cf0ccf2590f2f4"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"rYIPrEpdYe6SLAdSeOi1J01L+E4FL8cOoFbvW03bnFOMfyK5UXUWe7WDsUZGA9fWoxo/Ptovw9xOATBaNX/mBQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-09T05:05:31.022752Z","bundle_sha256":"ddcc47180c7ec598c1df6c70aab06ac9a8e3eb5347083d7d1d14f8b6a743373d"}}