ERIQ Embodied Reasoning Benchmark
AgiBot真实机器人试验图像构成的具身推理诊断集，含6,052组问答、15个子任务、4个推理维度和100多个场景；固定Hugging Face revision公开非gated，托管约4.27GB。它是VLM/VLA推理评测集而不是动作轨迹训练库，且Hugging Face数据卡的CC-BY-SA-4.0与GitHub README声称的CC-BY-4.0不一致。
- 开放状态
- 数据可下载·许可证冲突
- 许可证
- Hugging Face card: CC-BY-SA-4.0; GitHub README: CC-BY-4.0, but claimed LICENSE-DATA file is absent
- 商业使用
- not asserted until authors resolve license conflict; preserve attribution and share-alike if using hosted card terms
- 人工评审结论
- verified_public_real_robot_reasoning_benchmark_with_license_conflict
- 机器人
- AgiBot real-robot trials
- 任务
- spatial perception and grounding、planning and monitoring、error detection and recovery、human intent understanding
- 模态
- real-robot RGB images、single/sequential images、question、multiple-choice/binary answer、reasoning category
- 格式
- JSON question-answer files plus JPEG imagery
- 采集/生成方式
- curated from authentic first-person real-robot trials
- 已知问题
- License conflict; claimed license files absent at fixed GitHub commit; benchmark is not robot control data; no full-download checksum audit; no independent evaluation report.