{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NXIRZ2MARPXWQIETZZYOVDTQG7","short_pith_number":"pith:NXIRZ2MA","schema_version":"1.0","canonical_sha256":"6dd11ce9808bef682093ce70ea8e7037cfb93b4675739478477f7360f23f70b8","source":{"kind":"arxiv","id":"2602.17659","version":2},"attestation_state":"computed","paper":{"title":"When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.RO"],"primary_cat":"cs.CV","authors_text":"Daniel Szafir, Dong Jing, Jiaqi Liu, Mingyu Ding, Yuchun Feng, Yue Yang, Yu Fang, Zhenyu Wei","submitted_at":"2026-02-19T18:59:20Z","abstract_excerpt":"Vision-Language-Action models (VLAs) promise to ground language instructions in robot control, yet in practice often fail to faithfully follow language. When presented with instructions that lack strong scene-specific supervision, VLAs suffer from counterfactual failures: they act based on vision shortcuts induced by dataset biases, repeatedly executing well-learned behaviors and selecting objects frequently seen during training regardless of language intent. To systematically study it, we introduce LIBERO-CF, the first counterfactual benchmark for VLAs that evaluates language following capabi"},"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":"2602.17659","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-02-19T18:59:20Z","cross_cats_sorted":["cs.RO"],"title_canon_sha256":"f39356bf6e3747bf7dd276257c08a5249f673ebb056b3f7664f0ee3f6f0a4638","abstract_canon_sha256":"1d2351a04b59a90e932247e0d2d982eee3d4556d8779c92096eee3c6c391b90c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-16T01:22:35.108339Z","signature_b64":"DMXcpt4EHvdZ5XEldBvAlr65V/iGF5Tqhhm7xTeS1sYabna74CWanwBnNGBujrY4GvhEW99LlRPIAYjWtL8KCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6dd11ce9808bef682093ce70ea8e7037cfb93b4675739478477f7360f23f70b8","last_reissued_at":"2026-07-16T01:22:35.107340Z","signature_status":"signed_v1","first_computed_at":"2026-07-16T01:22:35.107340Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.RO"],"primary_cat":"cs.CV","authors_text":"Daniel Szafir, Dong Jing, Jiaqi Liu, Mingyu Ding, Yuchun Feng, Yue Yang, Yu Fang, Zhenyu Wei","submitted_at":"2026-02-19T18:59:20Z","abstract_excerpt":"Vision-Language-Action models (VLAs) promise to ground language instructions in robot control, yet in practice often fail to faithfully follow language. When presented with instructions that lack strong scene-specific supervision, VLAs suffer from counterfactual failures: they act based on vision shortcuts induced by dataset biases, repeatedly executing well-learned behaviors and selecting objects frequently seen during training regardless of language intent. To systematically study it, we introduce LIBERO-CF, the first counterfactual benchmark for VLAs that evaluates language following capabi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.17659","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/2602.17659/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":"2602.17659","created_at":"2026-07-16T01:22:35.107790+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.17659v2","created_at":"2026-07-16T01:22:35.107790+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.17659","created_at":"2026-07-16T01:22:35.107790+00:00"},{"alias_kind":"pith_short_12","alias_value":"NXIRZ2MARPXW","created_at":"2026-07-16T01:22:35.107790+00:00"},{"alias_kind":"pith_short_16","alias_value":"NXIRZ2MARPXWQIET","created_at":"2026-07-16T01:22:35.107790+00:00"},{"alias_kind":"pith_short_8","alias_value":"NXIRZ2MA","created_at":"2026-07-16T01:22:35.107790+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":9,"sample":[{"citing_arxiv_id":"2607.08127","citing_title":"Understanding and Mitigating the Video-Action Generalization Gap via Temporal Ratio","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2606.27295","citing_title":"LA4VLA: Learning to Act without Seeing via Language-Action Pretraining","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2606.23641","citing_title":"Flatness Preserves Instruction Following in Vision-Language-Action Models","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12366","citing_title":"APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09740","citing_title":"ProbeAct: Probe-Guided Training-Free Failure Recovery in Vision-Language-Action Models","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2606.02277","citing_title":"RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Models","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2606.27295","citing_title":"LA4VLA: Learning to Act without Seeing via Language-Action Pretraining","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17204","citing_title":"Event-Grounded Sparse Autoencoders for Vision-Language-Action Policies","ref_index":33,"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":6,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7","json":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7.json","graph_json":"https://pith.science/api/pith-number/NXIRZ2MARPXWQIETZZYOVDTQG7/graph.json","events_json":"https://pith.science/api/pith-number/NXIRZ2MARPXWQIETZZYOVDTQG7/events.json","paper":"https://pith.science/paper/NXIRZ2MA"},"agent_actions":{"view_html":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7","download_json":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7.json","view_paper":"https://pith.science/paper/NXIRZ2MA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.17659&json=true","fetch_graph":"https://pith.science/api/pith-number/NXIRZ2MARPXWQIETZZYOVDTQG7/graph.json","fetch_events":"https://pith.science/api/pith-number/NXIRZ2MARPXWQIETZZYOVDTQG7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7/action/storage_attestation","attest_author":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7/action/author_attestation","sign_citation":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7/action/citation_signature","submit_replication":"https://pith.science/pith/NXIRZ2MARPXWQIETZZYOVDTQG7/action/replication_record"}},"created_at":"2026-07-16T01:22:35.107790+00:00","updated_at":"2026-07-16T01:22:35.107790+00:00"}