{"paper":{"title":"SafeManip: A Property-Driven Benchmark for Temporal Safety Evaluation in Robotic Manipulation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A new benchmark shows that robotic manipulation policies often violate temporal safety rules even on tasks they complete successfully.","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Chengyue Huang, Khang Vo Huynh, Lu Feng, Sebastian Elbaum, Zsolt Kira","submitted_at":"2026-05-12T16:49:28Z","abstract_excerpt":"Robotic manipulation is typically evaluated by task success, but successful completion does not guarantee safe execution. Many safety failures are temporal: a robot may touch a clean surface after contamination or release an object before it is fully inside an enclosure. We introduce SafeManip, a property-driven benchmark to explicitly evaluate temporal safety properties in robotic manipulation, moving beyond prior evaluations that largely focus on task completion or per-state constraint violations. SafeManip defines reusable safety templates over finite executions using Linear Temporal Logic "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Results show that even strong models often behave unsafely. Task-success gains do not reliably translate into safer execution: many successful rollouts remain unsafe, while longer-horizon or more complex tasks expose more violations.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The LTLf safety templates and the mapping from observed rollouts to symbolic predicate traces accurately capture all relevant temporal safety properties without introducing false positives or missing critical real-world constraints.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SafeManip is a new benchmark that applies LTLf monitors to assess temporal safety properties across eight categories in robotic manipulation, demonstrating that task success frequently fails to ensure safe execution in vision-language-action policies.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A new benchmark shows that robotic manipulation policies often violate temporal safety rules even on tasks they complete successfully.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0a1876392ab6438af818d0cf487a218824735b37fcc5b8a4363a41c00fd095a0"},"source":{"id":"2605.12386","kind":"arxiv","version":2},"verdict":{"id":"850d389d-b100-40ee-8ba1-39779118c924","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T03:36:05.803212Z","strongest_claim":"Results show that even strong models often behave unsafely. Task-success gains do not reliably translate into safer execution: many successful rollouts remain unsafe, while longer-horizon or more complex tasks expose more violations.","one_line_summary":"SafeManip is a new benchmark that applies LTLf monitors to assess temporal safety properties across eight categories in robotic manipulation, demonstrating that task success frequently fails to ensure safe execution in vision-language-action policies.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The LTLf safety templates and the mapping from observed rollouts to symbolic predicate traces accurately capture all relevant temporal safety properties without introducing false positives or missing critical real-world constraints.","pith_extraction_headline":"A new benchmark shows that robotic manipulation policies often violate temporal safety rules even on tasks they complete successfully."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.12386/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-26T14:40:19.643466Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T13:31:25.053122Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-20T09:46:22.567775Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T22:41:58.221082Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"5887839b98fc958103d3559f742d458d8d827d2aa36504639bc88d0cb4f5c9e2"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"224f6469a73e6cb72e5afe6c57d5a00029b466606582210e56e7e21df4e900bc"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}