遇见数据集

svla_koch_pick_n_place_vla_steering_height_test4_fix_dataset

收藏
Hugging Face2026-06-08 更新2026-06-08 收录
官方服务:

资源简介:

该数据集是一个使用LeRobot创建的机器人数据集,聚焦于名为koch_follower的机器人类型。它包含100个episodes和27599帧数据,以parquet文件格式存储,总数据文件大小为100MB,视频文件大小为200MB,视频帧率为30fps。数据集仅提供训练分割(索引从0到100)。特征包括:动作(6个浮点值,表示肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪的位置)、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(前视、顶部和腕部摄像头,每个图像分辨率为480x640,3通道,视频编码为av1,无音频),以及时间戳、帧索引、episode索引、索引和任务索引等元数据。该数据集主要用于机器人任务,如模仿学习或强化学习。

This dataset is a robotics dataset developed using LeRobot, focusing on the robot platform named koch_follower. It contains 100 episodes and 27599 frames of data, stored in Parquet file format, with a total data file size of 100 MB and a video file size of 200 MB, at a frame rate of 30 fps. Only the training split is provided for this dataset, with indices ranging from 0 to 100. The dataset includes the following features: actions (6 floating-point values representing shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position), observation states (6 joint positions consistent with the actions), image observations from three cameras (front-facing, top-down, and wrist cameras; each image has a resolution of 480x640 with 3 channels, and the video is encoded in AV1 without audio), as well as metadata such as timestamps, frame index, episode index, index, and task index. This dataset is mainly intended for robotics tasks such as imitation learning or reinforcement learning.

提供机构:
ethanCSL
创建时间:
2026-06-08
原始信息汇总

数据集概述

  • 数据集名称: ethanCSL/svla_koch_pick_n_place_vla_steering_height_test4_fix_dataset
  • 许可证: 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): 100
  • 总帧数 (Frames): 27,599
  • 总任务数 (Tasks): 1
  • 数据文件大小: 100 MB
  • 视频文件大小: 200 MB
  • 帧率 (FPS): 30
  • 分割: 全部 100 个片段用于训练(train: 0:100

机器人类型

  • 机器人: koch_follower

数据特征

动作 (Action): 6 维浮点数组,包含以下关节位置:

  • shoulder_pan.pos
  • shoulder_lift.pos
  • elbow_flex.pos
  • wrist_flex.pos
  • wrist_roll.pos
  • gripper.pos

观测状态 (Observation State): 与动作结构相同,同样是 6 维浮点数组,表示各关节位置。

观测图像 (Observation Images): 包含三个摄像头视角,均为视频数据,分辨率 480x640,3 通道,AV1 编码,30 FPS:

  • front (前置摄像头)
  • top (顶部摄像头)
  • wrist (腕部摄像头)

其它特征:

  • timestamp: 时间戳 (float32)
  • frame_index: 帧索引 (int64)
  • episode_index: 片段索引 (int64)
  • index: 索引 (int64)
  • task_index: 任务索引 (int64)

数据存储结构

  • 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
二维码
社区交流群
二维码
科研交流群
商业服务