{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:HBIG27LHAKA7RHFZV3HH5T3YOP","short_pith_number":"pith:HBIG27LH","canonical_record":{"source":{"id":"2604.05378","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-07T03:24:22Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"17b1d5b53f237661761fc44f75959a809747c7a7b51a92995c5b2b3dd9f06c61","abstract_canon_sha256":"8b54d6f83f80aa3bb70193520d7af13124c6edb1133215ef05014b372b4f72eb"},"schema_version":"1.0"},"canonical_sha256":"38506d7d670281f89cb9aece7ecf7873da581e1922badd52d9afa8425ecae91e","source":{"kind":"arxiv","id":"2604.05378","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.05378","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"arxiv_version","alias_value":"2604.05378v1","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.05378","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"pith_short_12","alias_value":"HBIG27LHAKA7","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"pith_short_16","alias_value":"HBIG27LHAKA7RHFZ","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"pith_short_8","alias_value":"HBIG27LH","created_at":"2026-05-28T01:04:39Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:HBIG27LHAKA7RHFZV3HH5T3YOP","target":"record","payload":{"canonical_record":{"source":{"id":"2604.05378","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-07T03:24:22Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"17b1d5b53f237661761fc44f75959a809747c7a7b51a92995c5b2b3dd9f06c61","abstract_canon_sha256":"8b54d6f83f80aa3bb70193520d7af13124c6edb1133215ef05014b372b4f72eb"},"schema_version":"1.0"},"canonical_sha256":"38506d7d670281f89cb9aece7ecf7873da581e1922badd52d9afa8425ecae91e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:39.534633Z","signature_b64":"7o8Lr7JuqT5H/G48UFF6r7bvS+bZJb9Rht3Yx8zwOtQTTYo0udKMZiY5y5cOD0To4x1roesRbIaJ280kLjqfDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"38506d7d670281f89cb9aece7ecf7873da581e1922badd52d9afa8425ecae91e","last_reissued_at":"2026-05-28T01:04:39.534098Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:39.534098Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.05378","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-28T01:04:39Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"i2q38xP26UuAsDQjsr6HU/8UE+fwE8dj60IOGy+4IuYwVddOhhb21iqwZKl4yItd4Nre/4Pdc2XN5FzMsmDeCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T03:47:56.468573Z"},"content_sha256":"a5385f46f0a6835227825053e383112fdbb081863d7637ec6f2bbd2f2217f217","schema_version":"1.0","event_id":"sha256:a5385f46f0a6835227825053e383112fdbb081863d7637ec6f2bbd2f2217f217"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:HBIG27LHAKA7RHFZV3HH5T3YOP","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"ICR-Drive: Instruction Counterfactual Robustness for End-to-End Language-Driven Autonomous Driving","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Minor instruction changes cause large performance drops and new failure modes in language-conditioned driving models.","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Can Cui, Kaiser Hamid, Nade Liang","submitted_at":"2026-04-07T03:24:22Z","abstract_excerpt":"Recent progress in vision-language-action (VLA) models has enabled language-conditioned driving agents to execute natural-language navigation commands in closed-loop simulation, yet standard evaluations largely assume instructions are precise and well-formed. In deployment, instructions vary in phrasing and specificity, may omit critical qualifiers, and can occasionally include misleading, authority-framed text, leaving instruction-level robustness under-measured. We introduce ICR-Drive, a diagnostic framework for instruction counterfactual robustness in end-to-end language-conditioned autonom"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on LMDrive and BEVDriver show that minor instruction changes can induce substantial performance drops and distinct failure modes, revealing a reliability gap for deploying embodied foundation models in safety-critical driving.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The four perturbation families (Paraphrase, Ambiguity, Noise, Misleading) and the CARLA simulator setup sufficiently represent the distribution of real-world human instructions and driving conditions that would be encountered in deployment.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ICR-Drive reveals substantial performance drops in end-to-end language-driven driving models when instructions are paraphrased, made ambiguous, noised, or misleading.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Minor instruction changes cause large performance drops and new failure modes in language-conditioned driving models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f0d97b331dd626550e4583d5029f35c948215d85d972a475020ca30534f63505"},"source":{"id":"2604.05378","kind":"arxiv","version":1},"verdict":{"id":"54326683-dc56-4024-aa32-7a87b7460a1d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T19:38:30.241618Z","strongest_claim":"Experiments on LMDrive and BEVDriver show that minor instruction changes can induce substantial performance drops and distinct failure modes, revealing a reliability gap for deploying embodied foundation models in safety-critical driving.","one_line_summary":"ICR-Drive reveals substantial performance drops in end-to-end language-driven driving models when instructions are paraphrased, made ambiguous, noised, or misleading.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The four perturbation families (Paraphrase, Ambiguity, Noise, Misleading) and the CARLA simulator setup sufficiently represent the distribution of real-world human instructions and driving conditions that would be encountered in deployment.","pith_extraction_headline":"Minor instruction changes cause large performance drops and new failure modes in language-conditioned driving models."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.05378/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":26,"sample":[{"doi":"","year":2022,"title":"RT-1: Robotics Transformer for Real-World Control at Scale","work_id":"e11bda85-8531-46bc-a07f-d0ade3643ab1","ref_index":1,"cited_arxiv_id":"2212.06817","is_internal_anchor":true},{"doi":"","year":2020,"title":"Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom","work_id":"171a5474-47d6-4c54-8886-72ed2a242c5d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Personalized autonomous driving with large language models: Field experiments","work_id":"78e20a2b-4861-49a5-8b49-d8e7aebc0ae4","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Carla: An open urban driv- ing simulator","work_id":"a9f2f8b6-78e3-4a3e-9d6e-cad3141d39db","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Benchmarking neu- ral network robustness to common corruptions and perturba- tions, 2019","work_id":"956c1450-e026-4d1e-9279-92d47b67b665","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":26,"snapshot_sha256":"340398f68b7015661769ad897d8de2e30d4b63f6abd76e3cefe977926736c8ac","internal_anchors":3},"formal_canon":{"evidence_count":1,"snapshot_sha256":"21f6cb151f302da0abe45e60896e548e1dd70922d9cfb9e54bb79d51ea4b1813"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"54326683-dc56-4024-aa32-7a87b7460a1d"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-28T01:04:39Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Ogzy9eYq4bc6YrUE3objdR7HGKrzmLrtNFaRnbGwZJSaOuF1RNH4dWFCjJiuxL3RuDNWMvcG8CKUj8RYG6DTCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T03:47:56.469158Z"},"content_sha256":"849746faf18a98a1af7e6c22ea9b2b66c832994c7080d0c18eb2a65dc16ce915","schema_version":"1.0","event_id":"sha256:849746faf18a98a1af7e6c22ea9b2b66c832994c7080d0c18eb2a65dc16ce915"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/HBIG27LHAKA7RHFZV3HH5T3YOP/bundle.json","state_url":"https://pith.science/pith/HBIG27LHAKA7RHFZV3HH5T3YOP/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/HBIG27LHAKA7RHFZV3HH5T3YOP/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":"2026-06-03T03:47:56Z","links":{"resolver":"https://pith.science/pith/HBIG27LHAKA7RHFZV3HH5T3YOP","bundle":"https://pith.science/pith/HBIG27LHAKA7RHFZV3HH5T3YOP/bundle.json","state":"https://pith.science/pith/HBIG27LHAKA7RHFZV3HH5T3YOP/state.json","well_known_bundle":"https://pith.science/.well-known/pith/HBIG27LHAKA7RHFZV3HH5T3YOP/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:HBIG27LHAKA7RHFZV3HH5T3YOP","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":"8b54d6f83f80aa3bb70193520d7af13124c6edb1133215ef05014b372b4f72eb","cross_cats_sorted":["cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-07T03:24:22Z","title_canon_sha256":"17b1d5b53f237661761fc44f75959a809747c7a7b51a92995c5b2b3dd9f06c61"},"schema_version":"1.0","source":{"id":"2604.05378","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.05378","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"arxiv_version","alias_value":"2604.05378v1","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.05378","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"pith_short_12","alias_value":"HBIG27LHAKA7","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"pith_short_16","alias_value":"HBIG27LHAKA7RHFZ","created_at":"2026-05-28T01:04:39Z"},{"alias_kind":"pith_short_8","alias_value":"HBIG27LH","created_at":"2026-05-28T01:04:39Z"}],"graph_snapshots":[{"event_id":"sha256:849746faf18a98a1af7e6c22ea9b2b66c832994c7080d0c18eb2a65dc16ce915","target":"graph","created_at":"2026-05-28T01:04:39Z","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":"Experiments on LMDrive and BEVDriver show that minor instruction changes can induce substantial performance drops and distinct failure modes, revealing a reliability gap for deploying embodied foundation models in safety-critical driving."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The four perturbation families (Paraphrase, Ambiguity, Noise, Misleading) and the CARLA simulator setup sufficiently represent the distribution of real-world human instructions and driving conditions that would be encountered in deployment."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"ICR-Drive reveals substantial performance drops in end-to-end language-driven driving models when instructions are paraphrased, made ambiguous, noised, or misleading."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Minor instruction changes cause large performance drops and new failure modes in language-conditioned driving models."}],"snapshot_sha256":"f0d97b331dd626550e4583d5029f35c948215d85d972a475020ca30534f63505"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"21f6cb151f302da0abe45e60896e548e1dd70922d9cfb9e54bb79d51ea4b1813"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.05378/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Recent progress in vision-language-action (VLA) models has enabled language-conditioned driving agents to execute natural-language navigation commands in closed-loop simulation, yet standard evaluations largely assume instructions are precise and well-formed. In deployment, instructions vary in phrasing and specificity, may omit critical qualifiers, and can occasionally include misleading, authority-framed text, leaving instruction-level robustness under-measured. We introduce ICR-Drive, a diagnostic framework for instruction counterfactual robustness in end-to-end language-conditioned autonom","authors_text":"Can Cui, Kaiser Hamid, Nade Liang","cross_cats":["cs.CV"],"headline":"Minor instruction changes cause large performance drops and new failure modes in language-conditioned driving models.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-07T03:24:22Z","title":"ICR-Drive: Instruction Counterfactual Robustness for End-to-End Language-Driven Autonomous Driving"},"references":{"count":26,"internal_anchors":3,"resolved_work":26,"sample":[{"cited_arxiv_id":"2212.06817","doi":"","is_internal_anchor":true,"ref_index":1,"title":"RT-1: Robotics Transformer for Real-World Control at Scale","work_id":"e11bda85-8531-46bc-a07f-d0ade3643ab1","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom","work_id":"171a5474-47d6-4c54-8886-72ed2a242c5d","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Personalized autonomous driving with large language models: Field experiments","work_id":"78e20a2b-4861-49a5-8b49-d8e7aebc0ae4","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Carla: An open urban driv- ing simulator","work_id":"a9f2f8b6-78e3-4a3e-9d6e-cad3141d39db","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Benchmarking neu- ral network robustness to common corruptions and perturba- tions, 2019","work_id":"956c1450-e026-4d1e-9279-92d47b67b665","year":2019}],"snapshot_sha256":"340398f68b7015661769ad897d8de2e30d4b63f6abd76e3cefe977926736c8ac"},"source":{"id":"2604.05378","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-10T19:38:30.241618Z","id":"54326683-dc56-4024-aa32-7a87b7460a1d","model_set":{"reader":"grok-4.3"},"one_line_summary":"ICR-Drive reveals substantial performance drops in end-to-end language-driven driving models when instructions are paraphrased, made ambiguous, noised, or misleading.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Minor instruction changes cause large performance drops and new failure modes in language-conditioned driving models.","strongest_claim":"Experiments on LMDrive and BEVDriver show that minor instruction changes can induce substantial performance drops and distinct failure modes, revealing a reliability gap for deploying embodied foundation models in safety-critical driving.","weakest_assumption":"The four perturbation families (Paraphrase, Ambiguity, Noise, Misleading) and the CARLA simulator setup sufficiently represent the distribution of real-world human instructions and driving conditions that would be encountered in deployment."}},"verdict_id":"54326683-dc56-4024-aa32-7a87b7460a1d"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a5385f46f0a6835227825053e383112fdbb081863d7637ec6f2bbd2f2217f217","target":"record","created_at":"2026-05-28T01:04:39Z","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":"8b54d6f83f80aa3bb70193520d7af13124c6edb1133215ef05014b372b4f72eb","cross_cats_sorted":["cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-07T03:24:22Z","title_canon_sha256":"17b1d5b53f237661761fc44f75959a809747c7a7b51a92995c5b2b3dd9f06c61"},"schema_version":"1.0","source":{"id":"2604.05378","kind":"arxiv","version":1}},"canonical_sha256":"38506d7d670281f89cb9aece7ecf7873da581e1922badd52d9afa8425ecae91e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"38506d7d670281f89cb9aece7ecf7873da581e1922badd52d9afa8425ecae91e","first_computed_at":"2026-05-28T01:04:39.534098Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-28T01:04:39.534098Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"7o8Lr7JuqT5H/G48UFF6r7bvS+bZJb9Rht3Yx8zwOtQTTYo0udKMZiY5y5cOD0To4x1roesRbIaJ280kLjqfDA==","signature_status":"signed_v1","signed_at":"2026-05-28T01:04:39.534633Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.05378","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a5385f46f0a6835227825053e383112fdbb081863d7637ec6f2bbd2f2217f217","sha256:849746faf18a98a1af7e6c22ea9b2b66c832994c7080d0c18eb2a65dc16ce915"],"state_sha256":"a4488dc622159760dcda414b7e520f2cc18b47bd5793b91c8afb5693c3b360cd"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"StNwlWS7McxlCZ2dldR85ENxvrgDWNword8VyqVZp8uFKeFua6HwNNtYf28vNBH22QYgB8KMd5k4NS6P5gXoDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-03T03:47:56.471776Z","bundle_sha256":"93abaa8d0d26a30e91af05bcce988d5ad5e3f9d30c456536ee5c5cf83caaf337"}}