RobotisAI/evButtonPush-260429-02-hdf5

evButtonPush 是一个用于机器人电梯按钮按压任务的多摄像头RGB演示数据集。该数据集使用ROBOTIS OMY 6-DoF机械臂在Isaac Sim 5.1 + Isaac Lab环境中通过脚本化教师（DLS-IK + 力反馈）自动收集。核心特点是每个演示强制使用100种电梯USD变体中的一种，确保50个演示对应50种独特的电梯（无重复）。数据集包含高分辨率图像（1920 × 1200），来自三个摄像头（手腕、顶部和底部摄像头），以及各种状态观测和动作数据。数据以HDF5格式存储，每个演示文件约1 GB。数据集包含50个演示（实际上传49个，因1个写入失败被排除），所有演示均成功完成按钮按压、LED点亮、门打开和返回初始位置的任务。数据采集频率为30 Hz，每演示包含约650至740帧。数据集还包含详细的观测和动作维度信息，如关节位置、末端执行器位置、按钮位置等。

evButtonPush is a multi-camera RGB demonstration dataset for a robotic elevator button pressing task. The dataset was collected using the ROBOTIS OMY 6-DoF manipulator in the Isaac Sim 5.1 + Isaac Lab environment with a scripted teacher (DLS-IK + force-feedback). The key feature is that each demonstration uses a distinct elevator USD variant from a pool of 100, ensuring 50 demonstrations correspond to 50 unique elevators (no duplicates). The dataset includes high-resolution images (1920 × 1200) from three cameras (wrist, top, and belly cameras), along with various state observations and action data. The data is stored in HDF5 format, with each demo file approximately 1 GB in size. The dataset contains 50 demonstrations (49 actually uploaded, as one was excluded due to bad write), all of which successfully completed the tasks of button pressing, LED lighting, door opening, and returning to the home position. The data collection frequency is 30 Hz, with each demo containing approximately 650 to 740 frames. The dataset also includes detailed observation and action dimensions, such as joint positions, end-effector positions, button positions, etc.