Voxel51/rcs_utn_green_box
收藏资源简介:
RCS UTN Green Box (FiftyOne)数据集是一个来自纽伦堡技术大学机器人控制堆栈(RCS)生态系统的分组视频数据集,专注于多视角机器人操作任务——拾取绿色盒子。每个episode作为一个组,包含五个同步摄像头视角的视频(包括广角侧视图、腕部摄像头、右侧视图、鸟瞰图和侧视图),以及每帧的密集机器人本体感知数据(如关节状态、末端执行器位姿、夹爪状态)和动作数据。数据集还提供相机内参和外参、每步奖励、成功标志等帧级字段。该数据集用于训练和评估视觉-语言-动作(VLA)策略,支持机器人学习的大规模应用。数据集包含143个episodes,总计715个视频样本,语言指令统一为pick the green box。
rcs_utn_green_box is a grouped FiftyOne video dataset of a multi-view robot manipulation task — pick the green box — collected with the Robot Control Stack (RCS) ecosystem from the University of Technology Nuremberg. Each episode is a group with one synchronized video per camera (five camera perspectives: side_wide, wrist, side_right, bird_eye, side), plus dense robot proprioception and action data on every frame, including joint states, end-effector poses, gripper state, actions, and tracked cube pose. This dataset provides the kind of multi-view, multi-modal trajectory data used to train and evaluate Vision-Language-Action (VLA) policies. It contains 143 episodes (groups) with 715 total video samples, and a uniform language instruction: pick the green box.




