Sensho/libero-lerobot-v2

LIBERO (LeRobot v2 + MP4) 是一个机器人学习数据集，主要特点包括：
- **机器人**：Franka Emika Panda（单臂）
- **任务**：40个
- **帧数**：273,465
- **状态（8维）**：`[eef_pos×3, eef_axis_angle×3, gripper_qpos×2]`
- **动作（7维）**：`[delta_pos×3, delta_axis_angle×3, gripper×1]`
- **摄像头**：agentview（第三人称）和 wrist（手眼视角），256×256 RGB
- **FPS**：10
- **任务套件**：4个LIBERO套件（libero-spatial, libero-object, libero-goal, libero-10）
- **总片段数**：1,693

该数据集是`physical-intelligence/libero`的重新编码版本，主要变化包括将每帧PNG图像编码为每片段H.264 MP4视频，摄像头键名更改为`observation.images.agentview`和`observation.images.wrist`，并添加了每帧的`annotation.task`字符串列。

LIBERO (LeRobot v2 + MP4) is a robot learning dataset with the following key features:
- **Robot**: Franka Emika Panda (single arm)
- **Episodes**: 1,693 across 4 LIBERO suites (libero-spatial, libero-object, libero-goal, libero-10)
- **Total frames**: 273,465
- **Tasks**: 40
- **FPS**: 10
- **State (8-dim)**: `[eef_pos×3, eef_axis_angle×3, gripper_qpos×2]`
- **Action (7-dim)**: `[delta_pos×3, delta_axis_angle×3, gripper×1]`
- **Cameras**: agentview (third-person) + wrist (eye-in-hand), 256×256 RGB

This dataset is a re-encoding of `physical-intelligence/libero` with per-frame PNG images encoded into per-episode H.264 MP4 videos, camera keys renamed to `observation.images.agentview` and `observation.images.wrist`, and a per-frame `annotation.task` string column added.