VITRA Teleoperation Dataset
VITRA TeleData 是 Microsoft 随人类视频预训练 VLA 公开的真机微调数据：Realman RM75 7DoF 机械臂、XHand 12DoF 灵巧手和 Intel RealSense D455 头部相机的同步遥操作 episode。公开数据卡为 MIT、约5.57GB；每条 episode 成对提供 HDF5 数值状态/动作和 MP4 RGB，并将外参/法兰变换按 episode 给出。它是人类视频→灵巧机器人 VLA 的真实锚点，但当前发布仅右臂，不能当作人形全身或多本体操作基准。
- 开放状态
- 已核验开放
- 许可证
- MIT declared by official Hugging Face dataset card
- 商业使用
- MIT applies to the released TeleData card; robot hardware, third-party drivers and any linked upstream resources retain their own terms
- 人工评审结论
- official_public_mit_real_robot_vla_finetuning_dataset
- 机器人
- Realman RM75 7DoF arm、XHand 12DoF dexterous hand、Intel RealSense D455 head-mounted RGB camera
- 任务
- real-world teleoperation demonstrations for VITRA robot fine-tuning and XHand deployment
- 模态
- head-mounted RGB MP4、HDF5 robot state、HDF5 target actions、language instruction、camera intrinsics、episode-specific kinematic/extrinsic transforms、arm and hand availability masks
- 格式
- paired HDF5 and MP4 files; documented HDF5 groups for metadata, kinematics, observations, state, action and masks
- 采集/生成方式
- real-world teleoperation
- 已知问题
- Exact episode count, scene/object inventory, failure distribution, clock-offset audit, full-package checksum, privacy/consent statement and independent reproduction are not published in the card. Current release contains only right-arm data; dual-arm fields are format placeholders. Do not transfer VITRA-1M upstream human-video rights to this MIT TeleData release or vice versa.