BeingBeyond/Being-H-EDU_SO101
收藏资源简介:
该数据集是用于机器人模仿学习的LeRobot格式SO101数据集,专门针对pick_cube_plate操作任务。任务指令为:将立方体捡起放入盘子中。 在每个情节中,SO101跟随臂观察桌面场景并执行拾取放置行为:抓取立方体,将其移动到盘子上方,然后放入盘子内。数据集可用于Being-H-EDU教程工作空间,对基于Being-H0.5的策略进行后训练,但不仅限于此,它也是标准的LeRobot风格机器人数据集,可被支持相同模式的其他模仿学习或机器人策略训练管道加载。数据集包括多个变体:原始合并数据集(210个情节)、质量过滤后数据集(189个情节)、修剪后推荐训练集(189个情节)以及LeRobot v3.0布局版本。数据模式包括机器人状态、动作、手腕和外部摄像头视频等核心特征。
This is the LeRobot-formatted SO101 dataset for robotic imitation learning, specifically tailored for the pick_cube_plate manipulation task. The task instruction is: "Pick up the cube and place it into the plate." In each episode, the SO101 robotic arm observes the tabletop scene and executes pick-and-place behaviors: grasp the cube, move it above the plate, then place it inside the plate. This dataset can be applied to the Being-H-EDU tutorial workspace for post-training of policies based on Being-H0.5, and it is not limited to this use case. As a standard LeRobot-style robotic dataset, it can also be loaded by other imitation learning or robotic policy training pipelines that support the same data schema. The dataset includes multiple variants: the original merged dataset (210 episodes), the quality-filtered dataset (189 episodes), the pruned recommended training set (189 episodes), and the LeRobot v3.0 layout version. The data modalities cover core features such as robot states, actions, wrist and external camera videos.




