so101_pick_and_place_overhead_high
收藏资源简介:
本数据集使用LeRobot工具创建,是一个面向机器人技术领域的演示数据集,核心内容为机器人执行拾取和放置任务的交互数据。数据集包含50个完整的episodes,总计16,622个时间步(帧),对应1个任务。数据以分块形式存储,总数据文件约100 MB,关联的视频文件约200 MB,视频帧率为30 fps。数据集仅包含训练集。每个数据样本包含多个结构化字段:动作(action)是一个5维浮点向量,表示机器人五个关节(肩部平移、肩部抬升、肘部弯曲、腕部弯曲、夹爪)的目标位置;观测(observation)包含两部分:状态(state)也是一个5维浮点向量,表示上述五个关节的当前实际位置;图像(images)包含两个固定视角的视频流:顶部(top)视角和侧面(side)视角,每个视频的分辨率均为640x480(宽x高),3通道(RGB),编码格式为AV1。此外,每个样本还包含时间戳(timestamp)、帧索引(frame_index)、episode索引(episode_index)、全局索引(index)和任务索引(task_index)。该数据集适用于机器人模仿学习、行为克隆、视觉运动策略学习等任务的研究与开发。
This dataset is created using the LeRobot tool and is a demonstration dataset for the field of robotics, focusing on interaction data for robot pick-and-place tasks. It contains 50 complete episodes, totaling 16,622 time steps (frames), corresponding to 1 task. The data is stored in chunks, with a total data file size of approximately 100 MB and associated video files of about 200 MB, with a video frame rate of 30 fps. The dataset only includes a training set. Each data sample includes multiple structured fields: action is a 5-dimensional floating-point vector representing the target positions of five robot joints (shoulder translation, shoulder lift, elbow bend, wrist bend, gripper); observation consists of two parts: state is also a 5-dimensional floating-point vector representing the current actual positions of the five joints; images include two fixed-view video streams: top view and side view, each with a resolution of 640x480 (width x height), 3 channels (RGB), encoded in AV1 format. Additionally, each sample contains a timestamp, frame index, episode index, global index, and task index. This dataset is suitable for research and development in tasks such as robot imitation learning, behavior cloning, and visual-motor policy learning.
数据集概述
- 数据集名称: so101_pick_and_place_overhead_high
- 许可证: Apache-2.0
- 任务类别: 机器人学 (robotics)
- 标签: LeRobot
数据集结构
- 总片段数: 50
- 总帧数: 16622
- 总任务数: 1
- 帧率: 30 FPS
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 数据路径格式:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频路径格式:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4 - 训练/测试划分: 所有50个片段用于训练
特征描述
- 动作 (action): 包含5个自由度(肩部旋转、肩部升降、肘部弯曲、腕部弯曲、夹爪位置),数据类型为float32,形状为[5]
- 观察状态 (observation.state): 与动作维度一致,包含同样的5个自由度,数据类型为float32,形状为[5]
- 观察图像 (observation.images.top 和 side): 两个摄像头视角(顶部和侧面),分辨率均为480×640像素,3通道RGB,使用AV1编码,30 FPS,数据类型为视频
- 时间戳 (timestamp): float32类型,形状[1]
- 帧索引 (frame_index): int64类型,形状[1]
- 片段索引 (episode_index): int64类型,形状[1]
- 全局索引 (index): int64类型,形状[1]
- 任务索引 (task_index): int64类型,形状[1]
机器人信息
- 机器人类型: so_follower
引用
- BibTeX: 暂无具体引用信息




