{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:VRGU3HWG3QNAITVKGOEM6L5T2U","short_pith_number":"pith:VRGU3HWG","canonical_record":{"source":{"id":"2605.10821","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-11T16:37:34Z","cross_cats_sorted":[],"title_canon_sha256":"3dc4f0e973ab47b318a0620ad2af510bea692c3f9909196bcc1e02b85d131c0c","abstract_canon_sha256":"2da6653a374e20fcd25c2b2683eb686d6d6add9183df6d39315c4dfca243f51d"},"schema_version":"1.0"},"canonical_sha256":"ac4d4d9ec6dc1a044eaa3388cf2fb3d51b95c6bac70f71c201525ac74ec21dde","source":{"kind":"arxiv","id":"2605.10821","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.10821","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"arxiv_version","alias_value":"2605.10821v2","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10821","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"pith_short_12","alias_value":"VRGU3HWG3QNA","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"pith_short_16","alias_value":"VRGU3HWG3QNAITVK","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"pith_short_8","alias_value":"VRGU3HWG","created_at":"2026-07-17T01:21:50Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:VRGU3HWG3QNAITVKGOEM6L5T2U","target":"record","payload":{"canonical_record":{"source":{"id":"2605.10821","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-11T16:37:34Z","cross_cats_sorted":[],"title_canon_sha256":"3dc4f0e973ab47b318a0620ad2af510bea692c3f9909196bcc1e02b85d131c0c","abstract_canon_sha256":"2da6653a374e20fcd25c2b2683eb686d6d6add9183df6d39315c4dfca243f51d"},"schema_version":"1.0"},"canonical_sha256":"ac4d4d9ec6dc1a044eaa3388cf2fb3d51b95c6bac70f71c201525ac74ec21dde","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-17T01:21:50.540816Z","signature_b64":"zdGNLlyKckRoWGaDoTjruprgFdFjpbFTFZEwS/K18SesNCGcZTjPFvs6Fhe6ZgseR7VXlCijOdfS5VhHpfTdDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac4d4d9ec6dc1a044eaa3388cf2fb3d51b95c6bac70f71c201525ac74ec21dde","last_reissued_at":"2026-07-17T01:21:50.539934Z","signature_status":"signed_v1","first_computed_at":"2026-07-17T01:21:50.539934Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.10821","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-17T01:21:50Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"G0zouX+dHO1HZ8Awsulw2pGym4Wf06W760tR/XoQaTtfIfVHslPeUF+e9Oz1AtmtbEbBFg7P09UTvHZjJh13CQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-11T08:54:07.110385Z"},"content_sha256":"51e64b0a31d267716f7e3b53036625d68b34a0233e125e62284b6cd25e70cf96","schema_version":"1.0","event_id":"sha256:51e64b0a31d267716f7e3b53036625d68b34a0233e125e62284b6cd25e70cf96"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:VRGU3HWG3QNAITVKGOEM6L5T2U","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"UniSteer: Unified Noise Steering for Efficient Human-Guided VLA Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"UniSteer inverts human corrective actions to noise targets to jointly steer RL updates in VLA adaptation.","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Bin Liang, Chuheng Zhang, Junjie Lu, Jun Yang, Kaixin Wang, Li Zhao, Min Xu, Xinyao Qin, Yuhua Jiang","submitted_at":"2026-05-11T16:37:34Z","abstract_excerpt":"Diffusion-based vision-language-action (VLA) models have emerged as strong priors for robotic manipulation, yet adapting them to real-world distributions remains challenging. In particular, on-robot reinforcement learning (RL) is expensive and time-consuming, so effective adaptation depends on efficient policy improvement within a limited budget of real-world interactions. Noise-space RL lowers the cost by keeping the pretrained VLA fixed as a denoising generator while updating only a lightweight actor that predicts the noise. However, its performance is still limited due to inefficient autono"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Real-world experiments on diverse manipulation tasks show that UniSteer adapts more efficiently than strong noise-space RL and action-space human-in-the-loop baselines, improving the success rate from 20% to 90% in 66 minutes on average across four real-world adaptation tasks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The approximate action-to-noise inversion of the frozen flow-matching decoder recovers sufficiently accurate noise targets from human corrective actions so that the resulting supervision improves rather than harms the joint RL-plus-human optimization of the noise actor.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"UniSteer unifies human corrective actions and noise-space RL for VLA adaptation by inverting actions to noise targets, raising success rates from 20% to 90% in 66 minutes across four real-world manipulation tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"UniSteer inverts human corrective actions to noise targets to jointly steer RL updates in VLA adaptation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bc3fa096f8624bb29aeeca6a946be61745d221fafa239bb33d87437f77fa68a4"},"source":{"id":"2605.10821","kind":"arxiv","version":2},"verdict":{"id":"dd56647d-153d-4bae-a5c6-bb867648a044","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T04:05:04.305864Z","strongest_claim":"Real-world experiments on diverse manipulation tasks show that UniSteer adapts more efficiently than strong noise-space RL and action-space human-in-the-loop baselines, improving the success rate from 20% to 90% in 66 minutes on average across four real-world adaptation tasks.","one_line_summary":"UniSteer unifies human corrective actions and noise-space RL for VLA adaptation by inverting actions to noise targets, raising success rates from 20% to 90% in 66 minutes across four real-world manipulation tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The approximate action-to-noise inversion of the frozen flow-matching decoder recovers sufficiently accurate noise targets from human corrective actions so that the resulting supervision improves rather than harms the joint RL-plus-human optimization of the noise actor.","pith_extraction_headline":"UniSteer inverts human corrective actions to noise targets to jointly steer RL updates in VLA adaptation."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10821/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T05:22:00.355576Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T14:35:04.965141Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:17.645156Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:58:26.314829Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"938dc17d9eb34a3e8b17b37fa88c920916e216589543d70d47c66712713eaeca"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5d0c255c2c595e70c75cb5047b483658d272cb1014f78f98260ed3014dc5cf86"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"dd56647d-153d-4bae-a5c6-bb867648a044"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-17T01:21:50Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"hz+K7vGI4tQfEdXvzjC2kEtl+mBEVmG60heL5bwaupkdLEeg/g5MyD/pBxapTFMcVGpyY+YvNPR1nU91tl/3DA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-11T08:54:07.111797Z"},"content_sha256":"809c75f0368a80093150cb42b2b1bff4cc87a72f261568aa26997d3e4ada075c","schema_version":"1.0","event_id":"sha256:809c75f0368a80093150cb42b2b1bff4cc87a72f261568aa26997d3e4ada075c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/VRGU3HWG3QNAITVKGOEM6L5T2U/bundle.json","state_url":"https://pith.science/pith/VRGU3HWG3QNAITVKGOEM6L5T2U/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/VRGU3HWG3QNAITVKGOEM6L5T2U/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":"2026-08-11T08:54:07Z","links":{"resolver":"https://pith.science/pith/VRGU3HWG3QNAITVKGOEM6L5T2U","bundle":"https://pith.science/pith/VRGU3HWG3QNAITVKGOEM6L5T2U/bundle.json","state":"https://pith.science/pith/VRGU3HWG3QNAITVKGOEM6L5T2U/state.json","well_known_bundle":"https://pith.science/.well-known/pith/VRGU3HWG3QNAITVKGOEM6L5T2U/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VRGU3HWG3QNAITVKGOEM6L5T2U","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":"2da6653a374e20fcd25c2b2683eb686d6d6add9183df6d39315c4dfca243f51d","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-11T16:37:34Z","title_canon_sha256":"3dc4f0e973ab47b318a0620ad2af510bea692c3f9909196bcc1e02b85d131c0c"},"schema_version":"1.0","source":{"id":"2605.10821","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.10821","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"arxiv_version","alias_value":"2605.10821v2","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10821","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"pith_short_12","alias_value":"VRGU3HWG3QNA","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"pith_short_16","alias_value":"VRGU3HWG3QNAITVK","created_at":"2026-07-17T01:21:50Z"},{"alias_kind":"pith_short_8","alias_value":"VRGU3HWG","created_at":"2026-07-17T01:21:50Z"}],"graph_snapshots":[{"event_id":"sha256:809c75f0368a80093150cb42b2b1bff4cc87a72f261568aa26997d3e4ada075c","target":"graph","created_at":"2026-07-17T01:21:50Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Real-world experiments on diverse manipulation tasks show that UniSteer adapts more efficiently than strong noise-space RL and action-space human-in-the-loop baselines, improving the success rate from 20% to 90% in 66 minutes on average across four real-world adaptation tasks."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The approximate action-to-noise inversion of the frozen flow-matching decoder recovers sufficiently accurate noise targets from human corrective actions so that the resulting supervision improves rather than harms the joint RL-plus-human optimization of the noise actor."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"UniSteer unifies human corrective actions and noise-space RL for VLA adaptation by inverting actions to noise targets, raising success rates from 20% to 90% in 66 minutes across four real-world manipulation tasks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"UniSteer inverts human corrective actions to noise targets to jointly steer RL updates in VLA adaptation."}],"snapshot_sha256":"bc3fa096f8624bb29aeeca6a946be61745d221fafa239bb33d87437f77fa68a4"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5d0c255c2c595e70c75cb5047b483658d272cb1014f78f98260ed3014dc5cf86"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T05:22:00.355576Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T14:35:04.965141Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:17.645156Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T08:58:26.314829Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.10821/integrity.json","findings":[],"snapshot_sha256":"938dc17d9eb34a3e8b17b37fa88c920916e216589543d70d47c66712713eaeca","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Diffusion-based vision-language-action (VLA) models have emerged as strong priors for robotic manipulation, yet adapting them to real-world distributions remains challenging. In particular, on-robot reinforcement learning (RL) is expensive and time-consuming, so effective adaptation depends on efficient policy improvement within a limited budget of real-world interactions. Noise-space RL lowers the cost by keeping the pretrained VLA fixed as a denoising generator while updating only a lightweight actor that predicts the noise. However, its performance is still limited due to inefficient autono","authors_text":"Bin Liang, Chuheng Zhang, Junjie Lu, Jun Yang, Kaixin Wang, Li Zhao, Min Xu, Xinyao Qin, Yuhua Jiang","cross_cats":[],"headline":"UniSteer inverts human corrective actions to noise targets to jointly steer RL updates in VLA adaptation.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-11T16:37:34Z","title":"UniSteer: Unified Noise Steering for Efficient Human-Guided VLA Adaptation"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.10821","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-12T04:05:04.305864Z","id":"dd56647d-153d-4bae-a5c6-bb867648a044","model_set":{"reader":"grok-4.3"},"one_line_summary":"UniSteer unifies human corrective actions and noise-space RL for VLA adaptation by inverting actions to noise targets, raising success rates from 20% to 90% in 66 minutes across four real-world manipulation tasks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"UniSteer inverts human corrective actions to noise targets to jointly steer RL updates in VLA adaptation.","strongest_claim":"Real-world experiments on diverse manipulation tasks show that UniSteer adapts more efficiently than strong noise-space RL and action-space human-in-the-loop baselines, improving the success rate from 20% to 90% in 66 minutes on average across four real-world adaptation tasks.","weakest_assumption":"The approximate action-to-noise inversion of the frozen flow-matching decoder recovers sufficiently accurate noise targets from human corrective actions so that the resulting supervision improves rather than harms the joint RL-plus-human optimization of the noise actor."}},"verdict_id":"dd56647d-153d-4bae-a5c6-bb867648a044"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:51e64b0a31d267716f7e3b53036625d68b34a0233e125e62284b6cd25e70cf96","target":"record","created_at":"2026-07-17T01:21:50Z","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":"2da6653a374e20fcd25c2b2683eb686d6d6add9183df6d39315c4dfca243f51d","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-11T16:37:34Z","title_canon_sha256":"3dc4f0e973ab47b318a0620ad2af510bea692c3f9909196bcc1e02b85d131c0c"},"schema_version":"1.0","source":{"id":"2605.10821","kind":"arxiv","version":2}},"canonical_sha256":"ac4d4d9ec6dc1a044eaa3388cf2fb3d51b95c6bac70f71c201525ac74ec21dde","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ac4d4d9ec6dc1a044eaa3388cf2fb3d51b95c6bac70f71c201525ac74ec21dde","first_computed_at":"2026-07-17T01:21:50.539934Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-17T01:21:50.539934Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"zdGNLlyKckRoWGaDoTjruprgFdFjpbFTFZEwS/K18SesNCGcZTjPFvs6Fhe6ZgseR7VXlCijOdfS5VhHpfTdDA==","signature_status":"signed_v1","signed_at":"2026-07-17T01:21:50.540816Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.10821","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:51e64b0a31d267716f7e3b53036625d68b34a0233e125e62284b6cd25e70cf96","sha256:809c75f0368a80093150cb42b2b1bff4cc87a72f261568aa26997d3e4ada075c"],"state_sha256":"72748a7a4c72e7741af2d7dbea3444fc3d951cf50e3eccce8d1d2b4532233c6c"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"6+jHFP/6OzLig/marTcS9CrGGDEuUcjq/lxf/r3/whoa2q7U0GBqWCL3KeWXGBU3Q5cAUla75ywc60I9mAvQDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-11T08:54:07.119101Z","bundle_sha256":"6823d1c21e74e55006fd8b2c9abe1258aa7a84ceb23d8b28f448c833223f8df1"}}