code_and_dataset · 暂缓：许可证冲突
审核码：hold_license_conflict; 官方关系：verified_official_but_license_conflicted; 最近核验：2026-07-20代码：A lightweight Python dataloader and HDF5-to-LeRobot conversion script are public.
权重：No paper policy checkpoint was verified.
数据：The full LeRobot-format dataset and lite H1/G1 datasets resolve without gating; the project also documents task-level downloads.
许可证：conflict:MIT-vs-Apache-2.0-vs-undisclosed；GitHub README says the dataset is MIT but the repository has no LICENSE file; the main Hugging Face dataset card declares Apache-2.0; the separate H1 and G1 lite dataset cards disclose no license.
依赖：Python 3.8 or newer、16 GB RAM recommended、LeRobot loader or project dataloader
硬件/传感器：Offline use needs storage on the order of 1 TB for the documented full LeRobot copy、Source data uses Unitree G1 and H1, RealSense D435, LiDAR and hand pressure sensing
实机证据：Official project documents 10.3k physical humanoid trajectories, more than 3 million frames and 260 tasks across G1/H1.
独立复现：No independent model-training reproduction with a fixed commit, hardware and comparable policy metrics was found.
限制/反面证据：License conflict blocks direct reviewed-artifact promotion despite verified downloads.；The project reports both about 500 GB task archives and about 1 TB LeRobot storage; users must distinguish packaging from unique information volume.；No independent downstream-policy replication is established.
当前建议：高潜待持续验证；Keep as a high-priority dataset milestone, but display a license-red-flag badge and block commercial/practice recommendation until the owner publishes one consistent artifact-scoped license.
重点标签：recency:2025、infrastructure:humanoid-dataset、practice:download-verified、license:red-flag、storage:very-high、reproduction:downstream-missing
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4 · 来源 5 · 来源 6