{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.02277", "kind": "paper", "url": "https://arxiv.org/abs/2606.02277"}, {"id": "gh:zgc-embodyai/robosemanticbench", "kind": "repo", "url": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench"}], "authors": ["Bin Yu", "Yao Zhang", "Haishan Liu", "Shijie Lian", "Yuliang Wei", "Xiaopeng Lin", "Zhaolong Shen", "Changti Wu", "Ruina Hu", "Bailing Wang", "Cong Huang", "Kai Chen"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.02277", "source_url": "https://arxiv.org/abs/2606.02277"}, "categories": ["Robotics & Embodied AI", "cs.RO"], "description": {"en": "RoboSemanticBench (RSB) is an embodied benchmark evaluating whether vision-language-action models use instruction semantics to select and grasp the correct physical target among candidate blocks in response to math or general-knowledge questions. It includes six suites with four- or ten-choice variants, procedural arithmetic, GSM8K-style, and MMLU-style questions, with diagnostic metrics separating task success from grasp success."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.02277", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.02277", "title": "RoboSemanticBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.02277", "modality": null, "name": "RoboSemanticBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_robosemanticbench_9bca3113", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "cb6813d46bdcde67740f6f6a9242a0b7c6d7a0f52cad7fad226a95bec018f6c3", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.02277"}, "publisher": null, "released": "2026-06-01", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.02277", "source_url": "https://arxiv.org/abs/2606.02277"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-02277", "source": "claire_radar", "source_benchmark_id": "2606.02277", "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": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench", "githubScope": "benchmark_repo", "githubStars": 4, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 3, "hfPaperUrl": "https://huggingface.co/papers/2606.02277", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "587e7e6271eb53d0d9a42c28e6a342f95702b75866ea73cb01aba2d2069c3b3b", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "RoboSemanticBench (RSB) is an embodied benchmark evaluating whether vision-language-action models use instruction semantics to select and grasp the correct physical target among candidate blocks in response to math or general-knowledge questions. It includes six suites with four- or ten-choice variants, procedural arithmetic, GSM8K-style, and MMLU-style questions, with diagnostic metrics separating task success from grasp success.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce RoboSemanticBench (RSB), an embodied benchmark for diagnosing semantic grounding in action prediction: whether post-trained VLA models can use complex instruction semantics to select and manipulate the correct physical target."}, "familyId": "bmf_85bece8521c2", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench", "url": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench", "url": "https://huggingface.co/papers/2606.02277"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench", "url": "https://arxiv.org/abs/2606.02277"}], "scope": "benchmark_repo", "url": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_robosemanticbench_9bca3113", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Robotics"], "links": {"code": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.02277", "paper": "https://arxiv.org/abs/2606.02277", "pdf": "https://arxiv.org/pdf/2606.02277", "project": null, "report": "https://arxiv.org/abs/2606.02277"}, "metrics": [], "motivation": "Vision-language-action (VLA) models are built on the premise that semantic understanding from pretrained language or vision-language backbones should guide robot action prediction.", "name": "RoboSemanticBench", "oneLine": "RoboSemanticBench (RSB) is an embodied benchmark evaluating whether vision-language-action models use instruction semantics to select and grasp the correct physical target among candidate blocks in response to math or general-knowledge questions. It includes six suites with four- or ten-choice variants, procedural arithmetic, GSM8K-style, and MMLU-style questions, with diagnostic metrics separating task success from grasp success.", "paperTitle": "RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Models", "primaryDomain": "Robotics & Embodied AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.02277", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "ZGC-EmbodyAI", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/ZGC-EmbodyAI/RoboSemanticBench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-01", "source": {"authors": ["Bin Yu", "Yao Zhang", "Haishan Liu", "Shijie Lian", "Yuliang Wei", "Xiaopeng Lin", "Zhaolong Shen", "Changti Wu", "Ruina Hu", "Bailing Wang", "Cong Huang", "Kai Chen"], "categories": ["cs.RO"], "id": "2606.02277", "title": "RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Models", "type": "arxiv", "url": "https://arxiv.org/abs/2606.02277"}, "sourceUpdatedAt": "2026-06-01T14:02:37Z", "topics": ["cs.RO"], "venueAttempts": [], "whyItMatters": "RSB addresses the evaluation gap of measuring whether VLA models actually ground instruction semantics in action prediction, separating low-level manipulation from semantic selection. It provides a controlled, repeatable protocol with held-out questions, useful for diagnosing model capabilities and guiding improvements in embodied semantic understanding."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
