{"record": {"aliases": ["MANIGUARD: A Benchmark and Data Suite for Specification-Grounded Safety Evaluation and Improvement of Robotic Manipulation"], "artifacts": [{"id": "arxiv:2608.17386", "kind": "paper", "url": "https://arxiv.org/abs/2608.17386"}], "categories": ["Robot manipulation", "Robotics", "Robotics & Embodied AI", "Safety"], "description": {"en": "ManiGuard-Bench evaluates 1,000 locked scenarios across six contact-rich household task families, with safety specifications checked by LTL$_f$-grounded automaton monitors and metrics separating safe success from engaged-and-safe behavior."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.17386", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.17386", "title": "ManiGuard-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2608.17386", "modality": null, "name": "ManiGuard-Bench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_maniguard-a-benchmark-and-data-suite-for-s_fde8045a", "origin_source": "arxiv", "recognition_confidence": "0.55", "record_sha256": "6570ab4859717ea2d08cb5ba9f5c9989ce388180fcd34cb5fb3d1f1983e78812", "relation": "evaluates_only", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-08-25T23:23:46.758372Z", "source_url": "https://arxiv.org/abs/2608.17386"}, "publisher": null, "released": "2026-08-18", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.17386", "source_url": "https://arxiv.org/abs/2608.17386"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-17386", "source": "claire_radar", "source_benchmark_id": "2608.17386", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["MANIGUARD: A Benchmark and Data Suite for Specification-Grounded Safety Evaluation and Improvement of Robotic Manipulation"], "annotation": "Unknown", "applicationDomains": ["Robotics & Embodied AI"], "area": "Robotics & Embodied AI", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 2, "hfPaperUrl": "https://huggingface.co/papers/2608.17386", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "stale"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark targets safety in foundation-model manipulation with large-scale rollouts and hardware validation, likely attracting attention from robotics and AI safety communities.", "score": 65}, "capabilities": ["Robot manipulation"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-25T23:23:46.758372Z", "inputHash": "02fbaa3f9373140b328e5afcac18c9e19cbf4f86db63191c68c1377d6e3e92c9", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "We introduce ManiGuard, a specification-grounded framework for evaluating and improving the safety of foundation-model manipulation, comprising the ManiGuard-Bench task suite and a paired safety-annotated trajectory-generation pipeline.", "canonicalNameSource": "abstract", "decisionReason": "The benchmark is formally named in the abstract, defines repeatable locked tasks with fixed safety specifications and monitor-based evaluation, and provides a public path through released demonstrations and a pipeline.", "model": "deepseek-v4-pro", "reviewedAt": "2026-08-25T23:23:46.758372Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "ManiGuard-Bench evaluates 1,000 locked scenarios across six contact-rich household task families, with safety specifications checked by LTL$_f$-grounded automaton monitors and metrics separating safe success from engaged-and-safe behavior.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "score_submission", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "named entity is a model, method, framework, or dataset used in evaluation"], "snippet": "We introduce ManiGuard, a specification-grounded framework for evaluating and improving the safety of foundation-model manipulation, comprising the ManiGuard-Bench task suite and a paired safety-annotated trajectory-generation pipeline."}, "familyId": "bmf_b025643cd9d4", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_maniguard-a-benchmark-and-data-suite-for-s_fde8045a", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Robotics"], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2608.17386", "pdf": "https://arxiv.org/pdf/2608.17386", "project": null, "report": "https://arxiv.org/abs/2608.17386"}, "metrics": [], "motivation": "Foundation-model policies for robotic manipulation are advancing rapidly on task success, but rigorous evaluation of whether they succeed safely is still lacking.", "name": "ManiGuard-Bench", "oneLine": "ManiGuard-Bench evaluates 1,000 locked scenarios across six contact-rich household task families, with safety specifications checked by LTL$_f$-grounded automaton monitors and metrics separating safe success from engaged-and-safe behavior.", "paperTitle": "MANIGUARD: A Benchmark and Data Suite for Specification-Grounded Safety Evaluation and Improvement of Robotic Manipulation", "primaryDomain": "Robotics & Embodied AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.17386", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "ranking": {"90d": {"confidence": "Low", "coverage": 0.15, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": null, "value": null}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.19411886932144953, "value": 2}}, "confidence": "Low", "coverage": 0.15, "rank": null, "score": 45}, "method": "attention-ranking-v13", "rank": null, "score": 45}}, "readiness": "Paper only", "recognitionConfidence": 0.55, "relation": "evaluates_only", "releasedAt": "2026-08-18", "source": {"authors": ["Yiyan Peng", "Philip Wang", "Simon Sinong Zhan", "Yiqi Lyu", "Zhenyang Ni", "Jixin Yan", "Fiorelli Wong", "Ruochen Jiao", "Hang Yin", "Xinyu Cao", "Huajie Shao", "Manling Li", "Ruohan Zhang", "Qi Zhu"], "categories": ["cs.RO"], "id": "2608.17386", "title": "MANIGUARD: A Benchmark and Data Suite for Specification-Grounded Safety Evaluation and Improvement of Robotic Manipulation", "type": "arxiv", "url": "https://arxiv.org/abs/2608.17386"}, "sourceUpdatedAt": "2026-08-18T05:27:35Z", "topics": ["Robotics", "Safety"], "venueAttempts": [], "whyItMatters": "ManiGuard-Bench fills a gap by evaluating robotic manipulation safety independently of task success, revealing that up to 21% of successful rollouts violate safety specifications and enabling safety-aware fine-tuning with 8,000 demonstrations."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
