{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2609.10895", "kind": "paper", "url": "https://arxiv.org/abs/2609.10895"}, {"id": "hf:alan123/reacthuman-benchmark-scaled", "kind": "dataset", "url": "https://huggingface.co/datasets/Alan123/reacthuman-benchmark-scaled"}], "authors": ["Yizhan Li", "Jianxin You", "Mengyang Xiong", "Yinhuan Chen", "Zicheng Zhao", "Dekun Wu", "Dongqing Zhang", "Bang Liu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2609.10895", "source_url": "https://arxiv.org/abs/2609.10895"}, "categories": ["Multimodal", "Robotics & Embodied AI"], "description": {"en": "Evaluates human-like reactive decision-making in multimodal LLMs by simulating 1,000+ reproducible household hazard scenes with physics-grounded ground truth across 17 event families."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2609.10895", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2609.10895", "title": "ReactHuman"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2609.10895", "modality": null, "name": "ReactHuman", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "High", "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_reacthuman_dbe6f627", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "b49a16fdf9f86ebee56cd6bf8597e0327ef96ce68f71b55747635bf577a8d79b", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-11T00:52:53.785364Z", "source_url": "https://arxiv.org/abs/2609.10895"}, "publisher": null, "released": "2026-09-09", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2609.10895", "source_url": "https://arxiv.org/abs/2609.10895"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2609-10895", "source": "claire_radar", "source_benchmark_id": "2609.10895", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "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": "2026-09-14T00:00:00.000Z", "hfDataset": "Alan123/reacthuman-benchmark-scaled", "hfDatasetDownloads": 728, "hfDatasetLikes": 1, "hfPaperUpvotes": 54, "hfPaperUrl": "https://huggingface.co/papers/2609.10895", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetLikes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfPaperUpvotes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Physics-grounded embodied AI is a hot topic and the reproducible simulation setup may attract robotics and MLLM researchers.", "score": 58}, "capabilities": [], "confidence": "High", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-11T00:52:53.785364Z", "inputHash": "4d65f7a4af0f759338e9aba3bbe0375313d27ac2a190775825deca0ce50c1a9b", "model": "deepseek-v4-pro"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "We introduce ReactHuman, the first physics-grounded benchmark for human-like reactive decision-making, in which the evaluated MLLM acts as the brain of a simulated humanoid", "canonicalNameSource": "abstract", "decisionReason": "Public dataset with detailed metadata and ground truth on Hugging Face, clear task and scoring.", "evaluationEvidence": {"reuse": "https://huggingface.co/datasets/Alan123/reacthuman-benchmark-scaled", "scoring": "with exact, annotation-free ground truth derived from 240 Hz rigid-body simulation, including adversarial objects whose appearance contradicts their physics", "task": "the evaluated MLLM acts as the brain of a simulated humanoid facing sudden household hazards; it spans 17 event families and over 1,000 bit-for-bit reproducible scenes"}, "model": "deepseek-v4-pro", "reviewedAt": "2026-09-11T00:52:53.785364Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Evaluates human-like reactive decision-making in multimodal LLMs by simulating 1,000+ reproducible household hazard scenes with physics-grounded ground truth across 17 event families.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce ReactHuman, the first physics-grounded benchmark for human-like reactive decision-making, in which the evaluated MLLM acts as the brain of a simulated humanoid facing sudden household hazards; it spans 17 event families and over 1,000 bit-for-bit reproducible scenes with exact, annotation-free ground truth derived from 240 Hz rigid-body simulation, including adversarial objects whose appearance contradicts their physics (a foam anvil, a steel apple)."}, "familyId": "bmf_46a0990402c2", "firstSeenAt": "2026-09-11", "heat": null, "id": "bm_reacthuman_dbe6f627", "indexedAt": "2026-09-11T00:48:15.671466Z", "industrySectors": ["Robotics"], "links": {"code": null, "data": "https://huggingface.co/datasets/Alan123/reacthuman-benchmark-scaled", "paper": "https://arxiv.org/abs/2609.10895", "pdf": "https://arxiv.org/pdf/2609.10895", "project": null, "report": "https://arxiv.org/abs/2609.10895"}, "metrics": [], "motivation": "Reacting to sudden physical hazards (catching a slipping plate, dodging a falling knife) is both a meaningful test of embodied intelligence and a hard requirement for deploying multimodal large language models (MLLMs) as the decision coreof household robots.", "name": "ReactHuman", "oneLine": "Evaluates human-like reactive decision-making in multimodal LLMs by simulating 1,000+ reproducible household hazard scenes with physics-grounded ground truth across 17 event families.", "paperTitle": "ReactHuman: A Physics-Grounded Benchmark for Human-Like Reactive Decision-Making in Embodied Multimodal LLMs", "primaryDomain": "Robotics & Embodied AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2609.10895", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-09-11T00:48:15.671466Z"}, "publications": [], "publishers": [{"name": "Alan123", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://huggingface.co/datasets/Alan123/reacthuman-benchmark-scaled"}], "ranking": {"30d": {"confidence": "Medium", "coverage": 0.45, "datasetDownloadRank": 11, "datasetRankPopulation": 34, "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": 30, "historyDate": "2026-09-05", "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": null, "value": null}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": 0.7937713513148983, "value": 728}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.7461558876701149, "value": 54}}, "confidence": "Medium", "coverage": 0.45, "rank": 7, "score": 62}, "method": "attention-ranking-v13", "rank": 7, "score": 62}, "90d": {"confidence": "Medium", "coverage": 0.45, "datasetDownloadRank": 24, "datasetRankPopulation": 85, "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": 0.6788100525951517, "value": 728}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.7080741722402608, "value": 54}}, "confidence": "Medium", "coverage": 0.45, "rank": 21, "score": 58}, "method": "attention-ranking-v13", "rank": 21, "score": 58}}, "readiness": "Inspectable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-09-09", "source": {"authors": ["Yizhan Li", "Jianxin You", "Mengyang Xiong", "Yinhuan Chen", "Zicheng Zhao", "Dekun Wu", "Dongqing Zhang", "Bang Liu"], "categories": ["cs.RO", "cs.AI"], "id": "2609.10895", "title": "ReactHuman: A Physics-Grounded Benchmark for Human-Like Reactive Decision-Making in Embodied Multimodal LLMs", "type": "arxiv", "url": "https://arxiv.org/abs/2609.10895"}, "sourceUpdatedAt": "2026-09-09T22:56:21Z", "topics": ["Multimodal"], "venueAttempts": [], "whyItMatters": "Provides a reproducible, annotation-free testbed for embodied AI safety and physical reasoning, addressing the gap in physics-grounded evaluation."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
