{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HO4WRQX2PEVQJZ42GAJYE36FVA","short_pith_number":"pith:HO4WRQX2","schema_version":"1.0","canonical_sha256":"3bb968c2fa792b04e79a3013826fc5a824ce3800eaab9dfb55c9f9c7f7bb811c","source":{"kind":"arxiv","id":"2602.10503","version":2},"attestation_state":"computed","paper":{"title":"Towards Long-Lived Robots: Continual Learning VLA Models via Reinforcement Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Dongbin Zhao, Haoran Li, Shuai Tian, Yongzhen Huang, Yuan Liu, Yuhui Chen, Yupeng Zheng, Yuxing Qin","submitted_at":"2026-02-11T04:05:03Z","abstract_excerpt":"Pretrained on large-scale and diverse datasets, VLA models demonstrate strong generalization and adaptability as general-purpose robotic policies. However, Supervised Fine-Tuning (SFT), which serves as the primary mechanism for adapting VLAs to downstream domains, requires substantial amounts of task-specific data and is prone to catastrophic forgetting. To address these limitations, we propose LifeLong-RFT, a simple yet effective Reinforcement Fine-Tuning (RFT) strategy for VLA models independent of online environmental feedback and pre-trained reward models. By integrating chunking-level on-"},"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":"2602.10503","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-02-11T04:05:03Z","cross_cats_sorted":[],"title_canon_sha256":"9deaaa36139efaecac1e69be0667fbb0cf91edbe8fae948ac9e18d018eff6f67","abstract_canon_sha256":"47ec9095821d94b8a0b8731c7409c319931a8d971fce5c4124487700dd177628"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:05.549223Z","signature_b64":"6I600EDmUDmbg8kaCOX1hdGj2zvsqFGLm563kIYvBcNsFJy7cDBQFlZ1tXHz1L9oO1BiqHupGW6n/VasIvrDAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3bb968c2fa792b04e79a3013826fc5a824ce3800eaab9dfb55c9f9c7f7bb811c","last_reissued_at":"2026-05-20T00:03:05.548408Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:05.548408Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Long-Lived Robots: Continual Learning VLA Models via Reinforcement Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Dongbin Zhao, Haoran Li, Shuai Tian, Yongzhen Huang, Yuan Liu, Yuhui Chen, Yupeng Zheng, Yuxing Qin","submitted_at":"2026-02-11T04:05:03Z","abstract_excerpt":"Pretrained on large-scale and diverse datasets, VLA models demonstrate strong generalization and adaptability as general-purpose robotic policies. However, Supervised Fine-Tuning (SFT), which serves as the primary mechanism for adapting VLAs to downstream domains, requires substantial amounts of task-specific data and is prone to catastrophic forgetting. To address these limitations, we propose LifeLong-RFT, a simple yet effective Reinforcement Fine-Tuning (RFT) strategy for VLA models independent of online environmental feedback and pre-trained reward models. By integrating chunking-level on-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.10503","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/2602.10503/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":"2602.10503","created_at":"2026-05-20T00:03:05.548536+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.10503v2","created_at":"2026-05-20T00:03:05.548536+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.10503","created_at":"2026-05-20T00:03:05.548536+00:00"},{"alias_kind":"pith_short_12","alias_value":"HO4WRQX2PEVQ","created_at":"2026-05-20T00:03:05.548536+00:00"},{"alias_kind":"pith_short_16","alias_value":"HO4WRQX2PEVQJZ42","created_at":"2026-05-20T00:03:05.548536+00:00"},{"alias_kind":"pith_short_8","alias_value":"HO4WRQX2","created_at":"2026-05-20T00:03:05.548536+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13119","citing_title":"Towards Long-horizon Embodied Agents with Tool-Aligned Vision-Language-Action Models","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05711","citing_title":"Closing the Loop: Unified 3D Scene Generation and Immersive Interaction via LLM-RL Coupling","ref_index":2,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA","json":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA.json","graph_json":"https://pith.science/api/pith-number/HO4WRQX2PEVQJZ42GAJYE36FVA/graph.json","events_json":"https://pith.science/api/pith-number/HO4WRQX2PEVQJZ42GAJYE36FVA/events.json","paper":"https://pith.science/paper/HO4WRQX2"},"agent_actions":{"view_html":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA","download_json":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA.json","view_paper":"https://pith.science/paper/HO4WRQX2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.10503&json=true","fetch_graph":"https://pith.science/api/pith-number/HO4WRQX2PEVQJZ42GAJYE36FVA/graph.json","fetch_events":"https://pith.science/api/pith-number/HO4WRQX2PEVQJZ42GAJYE36FVA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA/action/storage_attestation","attest_author":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA/action/author_attestation","sign_citation":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA/action/citation_signature","submit_replication":"https://pith.science/pith/HO4WRQX2PEVQJZ42GAJYE36FVA/action/replication_record"}},"created_at":"2026-05-20T00:03:05.548536+00:00","updated_at":"2026-05-20T00:03:05.548536+00:00"}