{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:RE2WLRMIB6W73PK2LHATYHJB7O","short_pith_number":"pith:RE2WLRMI","schema_version":"1.0","canonical_sha256":"893565c5880fadfdbd5a59c13c1d21fb908f22c45575bdb2b3b852d144fe4054","source":{"kind":"arxiv","id":"2605.15157","version":1},"attestation_state":"computed","paper":{"title":"Hand-in-the-Loop: Improving Dexterous VLA via Seamless Interventional Correction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"HandITL blends human corrective intent with ongoing VLA policy execution to eliminate gesture jumps during dexterous hand takeovers.","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Liqun Huang, Nie Lin, Ruoshi Wen, Wei Xu, Xiao Ma, Xinjun Sheng, Zhengming Zhu, Zhuohang Li","submitted_at":"2026-05-14T17:51:40Z","abstract_excerpt":"Vision-Language-Action (VLA) models are prone to compounding errors in dexterous manipulation, where high-dimensional action spaces and contact-rich dynamics amplify small policy deviations over long horizons. While Interactive Imitation Learning (IIL) can refine policies through human takeover data, applying it to high-degree-of-freedom (DoF) robotic hands remains challenging due to a command mismatch between human teleoperation and policy execution at the takeover moment, which causes abrupt robot-hand configuration changes, or \"gesture jumps\". We present Hand-in-the-Loop (HandITL), a seamle"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15157","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-14T17:51:40Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"c048f0a39f28d3aff1857ff1755136452a5c20b4481281a8372b6ab987fe90a5","abstract_canon_sha256":"8adf952441590f81f4c6b6d6af7f722a24f6bdc18c5e0b873c219eb7f4c2ec3e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.2","canonical_sha256":"893565c5880fadfdbd5a59c13c1d21fb908f22c45575bdb2b3b852d144fe4054","last_reissued_at":"2026-05-17T21:57:18.749021Z","signature_status":"unsigned_v0","first_computed_at":"2026-05-17T21:40:25.426639Z"},"graph_snapshot":{"paper":{"title":"Hand-in-the-Loop: Improving Dexterous VLA via Seamless Interventional Correction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"HandITL blends human corrective intent with ongoing VLA policy execution to eliminate gesture jumps during dexterous hand takeovers.","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Liqun Huang, Nie Lin, Ruoshi Wen, Wei Xu, Xiao Ma, Xinjun Sheng, Zhengming Zhu, Zhuohang Li","submitted_at":"2026-05-14T17:51:40Z","abstract_excerpt":"Vision-Language-Action (VLA) models are prone to compounding errors in dexterous manipulation, where high-dimensional action spaces and contact-rich dynamics amplify small policy deviations over long horizons. While Interactive Imitation Learning (IIL) can refine policies through human takeover data, applying it to high-degree-of-freedom (DoF) robotic hands remains challenging due to a command mismatch between human teleoperation and policy execution at the takeover moment, which causes abrupt robot-hand configuration changes, or \"gesture jumps\". We present Hand-in-the-Loop (HandITL), a seamle"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"HandITL reduces takeover jitter by 99.8% and grasp failures by 87.5%, mean completion time by 19.1%, and produces policies that outperform those trained with standard teleoperation data by 19% on average across three long-horizon dexterous tasks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the seamless blending of human corrective intent with autonomous policy execution can be achieved without introducing new instabilities or losing the benefit of the human correction in high-dimensional action spaces and contact-rich dynamics.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HandITL blends human intent with policy execution to eliminate gesture jumps in dexterous VLA interventions, cutting jitter by 99.8%, grasp failures by 87.5%, and yielding 19% better refined policies.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"HandITL blends human corrective intent with ongoing VLA policy execution to eliminate gesture jumps during dexterous hand takeovers.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ec41bb2ceeee400f81d027d6c85984508cf4993011ae5ece556bdfd02f3c2099"},"source":{"id":"2605.15157","kind":"arxiv","version":1},"verdict":{"id":"acab7a65-81fe-4604-9b9a-4c59c6f5117d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:05:35.135590Z","strongest_claim":"HandITL reduces takeover jitter by 99.8% and grasp failures by 87.5%, mean completion time by 19.1%, and produces policies that outperform those trained with standard teleoperation data by 19% on average across three long-horizon dexterous tasks.","one_line_summary":"HandITL blends human intent with policy execution to eliminate gesture jumps in dexterous VLA interventions, cutting jitter by 99.8%, grasp failures by 87.5%, and yielding 19% better refined policies.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the seamless blending of human corrective intent with autonomous policy execution can be achieved without introducing new instabilities or losing the benefit of the human correction in high-dimensional action spaces and contact-rich dynamics.","pith_extraction_headline":"HandITL blends human corrective intent with ongoing VLA policy execution to eliminate gesture jumps during dexterous hand takeovers."},"references":{"count":34,"sample":[{"doi":"","year":2019,"title":"Sample efficient interactive end-to-end deep learning for self-driving cars with selective multi-class safe dataset aggregation","work_id":"acd7b7a4-c530-4a71-b0c8-784b3b16428b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"GR00T N1: An Open Foundation Model for Generalist Humanoid Robots","work_id":"e2db69c7-ee8a-4cb7-a761-7b8de1dfcf97","ref_index":2,"cited_arxiv_id":"2503.14734","is_internal_anchor":true},{"doi":"","year":2024,"title":"$\\pi_0$: A Vision-Language-Action Flow Model for General Robot Control","work_id":"f790abdc-a796-482f-a40d-f8ee035ecfc2","ref_index":3,"cited_arxiv_id":"2410.24164","is_internal_anchor":true},{"doi":"","year":2025,"title":"GR-3 Technical Report","work_id":"7f3b800b-0146-4c37-b178-0801cd4270db","ref_index":4,"cited_arxiv_id":"2507.15493","is_internal_anchor":true},{"doi":"","year":2025,"title":"Conrft: A reinforced fine-tuning method for vla models via consistency policy","work_id":"12e4ef8e-5311-42c5-86de-e8a0c0a4c1c1","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":34,"snapshot_sha256":"7413377d38a8bd29f190194c953a2e264ba0fd7cf5e887b4652ad54dbd1982e8","internal_anchors":6},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c5c069f49630d5d70a4d6eaab3e61864c2310c7360dbc4bfbe32bb29f1378c60"},"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":"2605.15157","created_at":"2026-05-17T21:18:33.054407+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15157v1","created_at":"2026-05-17T21:18:33.054407+00:00"},{"alias_kind":"pith_short_12","alias_value":"RE2WLRMIB6W7","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"RE2WLRMIB6W73PK2","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"RE2WLRMI","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O","json":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O.json","graph_json":"https://pith.science/api/pith-number/RE2WLRMIB6W73PK2LHATYHJB7O/graph.json","events_json":"https://pith.science/api/pith-number/RE2WLRMIB6W73PK2LHATYHJB7O/events.json","paper":"https://pith.science/paper/RE2WLRMI"},"agent_actions":{"view_html":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O","download_json":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O.json","view_paper":"https://pith.science/paper/RE2WLRMI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15157&json=true","fetch_graph":"https://pith.science/api/pith-number/RE2WLRMIB6W73PK2LHATYHJB7O/graph.json","fetch_events":"https://pith.science/api/pith-number/RE2WLRMIB6W73PK2LHATYHJB7O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O/action/storage_attestation","attest_author":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O/action/author_attestation","sign_citation":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O/action/citation_signature","submit_replication":"https://pith.science/pith/RE2WLRMIB6W73PK2LHATYHJB7O/action/replication_record"}},"created_at":"2026-05-17T21:18:33.054407+00:00","updated_at":"2026-05-17T21:57:18.749103+00:00"}