{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3ICG2E6RA6GZXUIXEZKABTNWL7","short_pith_number":"pith:3ICG2E6R","schema_version":"1.0","canonical_sha256":"da046d13d1078d9bd117265400cdb65fc4100408c09e0b4979a7aab90b2a236a","source":{"kind":"arxiv","id":"2605.13775","version":1},"attestation_state":"computed","paper":{"title":"RoboEvolve: Co-Evolving Planner-Simulator for Robotic Manipulation with Limited Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A co-evolutionary VLM-VGM loop on 500 unlabeled images raises planner success by 30 points and simulator success by 48 percent while beating fully supervised baselines.","cross_cats":["cs.CV"],"primary_cat":"cs.RO","authors_text":"Harold Haodong Chen, Sirui Chen, Wenhang Ge, Ying-Cong Chen, Yingjie Xu","submitted_at":"2026-05-13T16:54:36Z","abstract_excerpt":"The scalability of robotic manipulation is fundamentally bottlenecked by the scarcity of task-aligned physical interaction data. While vision-language models (VLMs) and video generation models (VGMs) hold promise for autonomous data synthesis, they suffer from semantic-spatial misalignment and physical hallucinations, respectively. To bridge this gap, we introduce RoboEvolve, a novel framework that couples a VLM planner and a VGM simulator into a mutually reinforcing co-evolutionary loop. Operating purely on unlabeled seed images, RoboEvolve leverages a cognitive-inspired dual-phase mechanism:"},"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":false},"canonical_record":{"source":{"id":"2605.13775","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2026-05-13T16:54:36Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"328a3b6e0c3d344bb503df46cd324d610552c22aaa5b03aef663bacb9a2258cf","abstract_canon_sha256":"cfd3e2b32bbeda56f2440d71939ec6e7c4400f2a89d85d9495f953b2210152f8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:15.926931Z","signature_b64":"F/zejwfet5NqbzbB2kcCVG5NgXIoqNK60pUaxW9/AeD6MSljK8CvuS/U9zpijEI1h7kM5bSQpXuIuZ3Ihn0TBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"da046d13d1078d9bd117265400cdb65fc4100408c09e0b4979a7aab90b2a236a","last_reissued_at":"2026-05-18T02:44:15.926422Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:15.926422Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RoboEvolve: Co-Evolving Planner-Simulator for Robotic Manipulation with Limited Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A co-evolutionary VLM-VGM loop on 500 unlabeled images raises planner success by 30 points and simulator success by 48 percent while beating fully supervised baselines.","cross_cats":["cs.CV"],"primary_cat":"cs.RO","authors_text":"Harold Haodong Chen, Sirui Chen, Wenhang Ge, Ying-Cong Chen, Yingjie Xu","submitted_at":"2026-05-13T16:54:36Z","abstract_excerpt":"The scalability of robotic manipulation is fundamentally bottlenecked by the scarcity of task-aligned physical interaction data. While vision-language models (VLMs) and video generation models (VGMs) hold promise for autonomous data synthesis, they suffer from semantic-spatial misalignment and physical hallucinations, respectively. To bridge this gap, we introduce RoboEvolve, a novel framework that couples a VLM planner and a VGM simulator into a mutually reinforcing co-evolutionary loop. Operating purely on unlabeled seed images, RoboEvolve leverages a cognitive-inspired dual-phase mechanism:"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"RoboEvolve elevates base planners by 30 absolute points and amplifies simulator success by 48% on average, surpassing fully supervised baselines with merely 500 unlabeled seeds—a 50x reduction—while demonstrating robust continual learning without catastrophic forgetting.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the semantic-controlled multi-granular reward and nighttime mining of near-miss failures can enforce physical grounding and eliminate hallucinations in the VGM without any external verification or real-robot feedback.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A co-evolutionary VLM-VGM loop on 500 unlabeled images raises planner success by 30 points and simulator success by 48 percent while beating fully supervised baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"e3005a924468904730e5c5282061ea2e1683f7edb2555c966557470a4f64ca05"},"source":{"id":"2605.13775","kind":"arxiv","version":1},"verdict":{"id":"698432ae-044f-499b-8d9c-ad4b92d3ffda","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T17:51:46.495579Z","strongest_claim":"RoboEvolve elevates base planners by 30 absolute points and amplifies simulator success by 48% on average, surpassing fully supervised baselines with merely 500 unlabeled seeds—a 50x reduction—while demonstrating robust continual learning without catastrophic forgetting.","one_line_summary":"A co-evolutionary VLM-VGM loop on 500 unlabeled images raises planner success by 30 points and simulator success by 48 percent while beating fully supervised baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the semantic-controlled multi-granular reward and nighttime mining of near-miss failures can enforce physical grounding and eliminate hallucinations in the VGM without any external verification or real-robot feedback.","pith_extraction_headline":""},"references":{"count":83,"sample":[{"doi":"","year":null,"title":"Mind-v: Hierarchical video generation for long-horizon robotic manipulation with rl-based physical alignment","work_id":"4a9bf4e3-cdb2-4f3d-a0d5-9a5d730eef6f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer","work_id":"f38fc088-12aa-4bf4-9ecd-08d3e797ccb7","ref_index":2,"cited_arxiv_id":"2408.06072","is_internal_anchor":true},{"doi":"","year":2024,"title":"European conference on computer vision , pages=","work_id":"4bc13a3e-6c33-434d-92e8-e61c499a776f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"HunyuanVideo: A Systematic Framework For Large Video Generative Models","work_id":"881efa7e-7e73-4c66-9cc3-2803e551061c","ref_index":4,"cited_arxiv_id":"2412.03603","is_internal_anchor":true},{"doi":"","year":null,"title":"World Action Models are Zero-shot Policies","work_id":"9a85fc69-74df-450e-94cd-69d186e9e830","ref_index":5,"cited_arxiv_id":"2602.15922","is_internal_anchor":true}],"resolved_work":83,"snapshot_sha256":"3855d36bd55a549ac6a762c270f2630dfcfe47942979559367263fc7be94b689","internal_anchors":27},"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":"2605.13775","created_at":"2026-05-18T02:44:15.926503+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.13775v1","created_at":"2026-05-18T02:44:15.926503+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13775","created_at":"2026-05-18T02:44:15.926503+00:00"},{"alias_kind":"pith_short_12","alias_value":"3ICG2E6RA6GZ","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"3ICG2E6RA6GZXUIX","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"3ICG2E6R","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":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7","json":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7.json","graph_json":"https://pith.science/api/pith-number/3ICG2E6RA6GZXUIXEZKABTNWL7/graph.json","events_json":"https://pith.science/api/pith-number/3ICG2E6RA6GZXUIXEZKABTNWL7/events.json","paper":"https://pith.science/paper/3ICG2E6R"},"agent_actions":{"view_html":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7","download_json":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7.json","view_paper":"https://pith.science/paper/3ICG2E6R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.13775&json=true","fetch_graph":"https://pith.science/api/pith-number/3ICG2E6RA6GZXUIXEZKABTNWL7/graph.json","fetch_events":"https://pith.science/api/pith-number/3ICG2E6RA6GZXUIXEZKABTNWL7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7/action/storage_attestation","attest_author":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7/action/author_attestation","sign_citation":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7/action/citation_signature","submit_replication":"https://pith.science/pith/3ICG2E6RA6GZXUIXEZKABTNWL7/action/replication_record"}},"created_at":"2026-05-18T02:44:15.926503+00:00","updated_at":"2026-05-18T02:44:15.926503+00:00"}