soarm101-feeding-nuts

该数据集是SO-ARM101机器人执行“喂食坚果”操作任务的远程操作演示数据，采用Apache TsFile格式进行存储。数据集包含80个独立的演示片段，总计记录了74,355个时间步，数据以每秒30帧的速率采集。每个时间步均包含6自由度关节位置观测（具体包括肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕旋转和夹爪状态）以及对应的6自由度关节位置动作命令。数据存储在单个TsFile文件中，使用episode_index（范围从0到79）和task_index（始终为0）作为标签维度来组织数据。需要注意的是，原始数据集中的视频文件（如手腕摄像头和侧面摄像头流）未包含在此版本中，但用户可从原始数据源获取这些视频。该数据集基于LeRobot v3.0框架创建，主要适用于机器人模仿学习（如行为克隆、强化学习）和时间序列分析（如动作预测、状态建模）等研究任务。

This dataset consists of teleoperated demonstration data for the feeding nuts manipulation task performed by the SO-ARM101 robot, stored in Apache TsFile format. It includes 80 demonstration episodes, totaling 74,355 timesteps recorded at a rate of 30 frames per second. Each timestep contains 6-degree-of-freedom joint position observations (specifically shoulder translation, shoulder lift, elbow bend, wrist bend, wrist rotation, and gripper) along with corresponding 6-degree-of-freedom joint position action commands. The data is stored in a single TsFile file, organized using episode_index (ranging from 0 to 79) and task_index (always 0) as label dimensions. Note that video files from the original dataset (such as wrist and side camera streams) are not included in this version but can be obtained from the original source. The dataset is built on LeRobot v3.0 and is suitable for robotics imitation learning (e.g., behavior cloning, reinforcement learning) and time series analysis tasks (e.g., action prediction, state modeling).