eval_LAVLA_Async_VI_100M_40k_b20_whiteblackcube_v2_2026_05_27_white_latency_bench_v2
收藏资源简介:
该数据集是一个机器人操作数据集,由LeRobot创建,专门针对so_follower类型的机器人。它包含2个完整的情节(episodes),总计3376帧,覆盖1个任务。数据以Parquet文件格式存储,总大小为100MB(数据文件)和200MB(视频文件),帧率为30fps。数据集的特征包括:动作数据(6个浮点值,表示机器人关节位置,如肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置)、观测状态(与动作相同的6个关节位置)、来自三个摄像头的图像观测(每个摄像头提供480x640分辨率、3通道的视频数据),以及时间戳、帧索引、情节索引等元数据。图像数据采用AV1编解码器,像素格式为yuv420p,无音频。该数据集适用于机器人控制、强化学习和计算机视觉任务,许可证为Apache 2.0。
This is a robotic manipulation dataset developed by LeRobot, specifically tailored for so_follower-type robots. It contains 2 complete episodes, totaling 3376 frames, covering 1 single task. The dataset is stored in Parquet file format, with a total size of 100MB for the data files and 200MB for the video files, with a frame rate of 30fps. The dataset includes the following features: action data (6 floating-point values representing robot joint positions, including shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation and gripper position), observation states (the same 6 joint positions as the action data), image observations from three cameras (each providing 3-channel video data with a resolution of 480x640), as well as metadata such as timestamps, frame indices and episode indices. The image data adopts the AV1 codec, with a pixel format of yuv420p, and contains no audio. This dataset is suitable for robotic control, reinforcement learning and computer vision tasks, and is licensed under Apache 2.0.
数据集概述
该数据集是一个用于机器人学习的异步视觉-语言-动作数据集,通过LeRobot框架创建,旨在为机器人操作任务提供训练数据。
- 许可协议:Apache-2.0
- 任务类别:机器人学 (Robotics)
- 标签:LeRobot
数据集规模
- 总片段数 (Episodes):2
- 总帧数 (Frames):3376
- 总任务数 (Tasks):1
- 帧率 (FPS):30
- 数据文件大小:100 MB
- 视频文件大小:200 MB
- 代码库版本:v3.0
- 机器人类型:so_follower
- 训练/测试拆分:仅包含训练集(所有数据作为训练集,索引 0:2)
数据格式
数据以 Parquet 文件存储,路径为 data/*/*.parquet。每个片段包含多个数据块,每个块包含多个文件。
- 数据路径:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
特征字段
每个样本包含以下特征:
-
action (动作)
- 类型:float32,形状 [6]
- 字段:
shoulder_pan.pos,shoulder_lift.pos,elbow_flex.pos,wrist_flex.pos,wrist_roll.pos,gripper.pos
-
observation.state (观测状态)
- 类型:float32,形状 [6]
- 字段:
shoulder_pan.pos,shoulder_lift.pos,elbow_flex.pos,wrist_flex.pos,wrist_roll.pos,gripper.pos
-
observation.images.camera1 / camera2 / camera3 (三路相机图像)
- 类型:视频 (AV1 编码)
- 分辨率:480x640,3 通道
- 帧率:30 FPS
- 是否为深度图:否
- 无音频
-
timestamp (时间戳):float32,形状 [1]
-
frame_index (帧索引):int64,形状 [1]
-
episode_index (片段索引):int64,形状 [1]
-
index (全局索引):int64,形状 [1]
-
task_index (任务索引):int64,形状 [1]
可视化
该数据集支持在 LeRobot 可视化工具中在线预览,可通过页面上的"Visualize this dataset"按钮访问。



