pick_orange_aligned_100
收藏资源简介:
本数据集为 SO101 PickOrange Aligned 100,由宇哥的具身笔记在 LeIsaac / Isaac Lab 仿真环境中生成,采用 LeRobot Dataset v3 格式。任务为 LeIsaac-SO101-PickOrange-v0,使用 SO101 机械臂(so101_follower)进行 PickOrange 任务。数据集包含 100 个 episodes,共 97,600 帧,帧率为 30 FPS。动作和状态维度均为 6 维。配备 front 和 wrist 两路相机,图像尺寸为 640×480,视频编码采用 AV1。数据划分中训练集为 0 到 100 号 episode。数据结构包括 data/(每帧状态、动作、时间戳和索引)、meta/(数据集说明、统计量、任务和 episode 元数据)、videos/(front 与 wrist 两路相机视频)。主要字段有:observation.state(SO101 六维关节状态)、action(SO101 六维关节动作)、observation.images.front(前视相机图像)、observation.images.wrist(腕部相机图像)、timestamp、frame_index、episode_index、task_index。该数据集适用于具身智能、机器人模仿学习研究以及 LeRobot 数据处理教学,可用于数据读取、可视化、质量检查及策略训练实验。注意:数据来自仿真环境,不代表真实机械臂上的传感器噪声、动力学误差和安全约束,将模型迁移到真机前需进行标定、安全限制和 sim-to-real 验证。
This dataset is SO101 PickOrange Aligned 100, generated by Yuges Embodied Notes in the LeIsaac / Isaac Lab simulation environment, using the LeRobot Dataset v3 format. The task is LeIsaac-SO101-PickOrange-v0, using the SO101 robotic arm (so101_follower) to perform the PickOrange task. The dataset contains 100 episodes, totaling 97,600 frames, with a frame rate of 30 FPS. Both action and state dimensions are 6. It is equipped with front and wrist cameras, image size is 640×480, and video encoding uses AV1. The training set includes episodes 0 to 100. The data structure includes data/ (state, action, timestamp, and index per frame), meta/ (dataset description, statistics, task and episode metadata), and videos/ (front and wrist camera videos). Main fields include: observation.state (SO101 6-DoF joint state), action (SO101 6-DoF joint action), observation.images.front (front camera image), observation.images.wrist (wrist camera image), timestamp, frame_index, episode_index, task_index. This dataset is suitable for embodied intelligence, robot imitation learning research, and LeRobot data processing teaching, and can be used for data reading, visualization, quality checking, and policy training experiments. Note: The data comes from a simulation environment and does not represent sensor noise, dynamic errors, or safety constraints of a real robotic arm. Before transferring the model to a real robot, calibration, safety limits, and sim-to-real verification are required.
数据集概述
- 名称:SO101 PickOrange Aligned 100
- 任务:LeIsaac-SO101-PickOrange-v0(SO101 机械臂抓取橙子任务)
- 机器人:so101_follower
- 数据格式:LeRobot Dataset v3
- 生成环境:LeIsaac / Isaac Lab 仿真
数据规模
- Episodes:100
- Frames:97,600
- FPS:30
- 数据划分:train
0:100(全部为训练数据)
数据维度
- 动作维度:6(六维关节动作)
- 状态维度:6(六维关节状态)
传感器与视觉
- 相机:front(前视)与 wrist(腕部)
- 图像尺寸:640 × 480
- 视频编码:AV1
- 图像字段:
observation.images.front、observation.images.wrist
数据结构
text
data/ 每帧状态、动作、时间戳和索引
meta/ 数据集说明、统计量、任务和 episode 元数据
videos/ front 与 wrist 两路相机视频
主要字段:
observation.state:SO101 六维关节状态action:SO101 六维关节动作observation.images.front:前视相机图像observation.images.wrist:腕部相机图像timestamp、frame_index、episode_index、task_index
用途与限制
- 用途:具身智能、机器人模仿学习、LeRobot 数据处理教学(数据读取、可视化、质量检查、策略训练)
- 限制:仿真数据,不含真实机械臂的传感器噪声、动力学误差和安全约束;迁移至真机前需进行标定、安全限制和 sim-to-real 验证
相关项目
配套代码:《具身智能从入门到实践》笔记学习代码




