QuadLocoManip
收藏资源简介:
QuadLocoManip 是一个用于机器人学,特别是四足机器人移动操作(Quadrupedal Loco-Manipulation)研究的模拟数据集。该数据集使用 LeRobot 代码库(版本 v3.0)生成,针对 B2Z1 型机器人。数据内容包含机器人执行任务时的观测、动作和状态信息:观测数据包括末端执行器的 8 维位姿(含抓取器状态)和来自腕部摄像头的 RGB 视频流(分辨率 480x640,帧率 50 fps);动作数据为 8 维向量,控制末端执行器的目标位姿和期望抓取器状态;状态数据提供了末端执行器在世界坐标系中的 7 维位姿。此外,数据集还包含时间戳、帧索引、episode 索引和任务索引等元数据。数据规模为 3500 个 episodes,总计 875,000 帧,覆盖 250 个不同的任务。数据以块的形式组织(每块 1000 帧),总数据文件大小约为 100 MB,视频文件大小约为 200 MB。数据集仅提供训练集划分,适用于机器人强化学习、模仿学习、移动操作策略学习等任务的研究与开发。
QuadLocoManip is a simulated dataset for robotics research, specifically focused on quadrupedal loco-manipulation. It is generated using the LeRobot codebase (version v3.0) for the B2Z1 robot. The dataset contains observations, actions, and state information during task execution: observations include an 8-dimensional end-effector pose (with gripper state) and RGB video streams from a wrist-mounted camera (resolution 480x640, frame rate 50 fps); actions are 8-dimensional vectors controlling the target pose and desired gripper state of the end-effector; state data provides a 7-dimensional end-effector pose in the world coordinate system. Additionally, the dataset includes metadata such as timestamps, frame indices, episode indices, and task indices. It comprises 3500 episodes, totaling 875,000 frames, covering 250 different tasks. Data is organized in chunks (1000 frames per chunk), with total data file size approximately 100 MB and video file size approximately 200 MB. The dataset only provides a training set split and is suitable for research and development in robot reinforcement learning, imitation learning, and mobile manipulation policy learning.
数据集概述:QuadLocoManip
- 数据集名称: QuadLocoManip
- 许可证: Apache-2.0
- 任务类别: 机器人学(Robotics)
- 标签: LeRobot, 四足运动操控(Quadrupedal Loco-Manip), 仿真数据(Simulated Data)
- 数据集创建工具: 使用 LeRobot 创建
数据集规模与结构
- 总集数(episodes): 3,500
- 总帧数(frames): 875,000
- 总任务数(tasks): 250
- 帧率(FPS): 50
- 数据文件大小: 100 MB(Parquet格式)
- 视频文件大小: 200 MB(MP4格式)
- 分块大小(chunks size): 1,000
- 数据划分: 仅训练集(train),范围为 0 至 3,500 集
数据文件格式
- 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
特征字段说明
| 字段名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
observation.ee_pose |
float32 | [8] | 末端执行器位姿:x, y, z, 四元数(w, x, y, z), 夹爪状态 |
observation.images.wrist |
video | [480, 640, 3] | 腕部摄像头视频,分辨率480x640,3通道,H264编码,帧率50 FPS |
action |
float32 | [8] | 动作指令:x, y, z, 四元数(w, x, y, z), 期望夹爪状态 |
state.ee_world |
float32 | [7] | 世界坐标系下的末端执行器状态:x, y, z, 四元数(w, x, y, z) |
timestamp |
float32 | [1] | 时间戳 |
frame_index |
int64 | [1] | 帧索引 |
episode_index |
int64 | [1] | 集索引 |
index |
int64 | [1] | 全局索引 |
task_index |
int64 | [1] | 任务索引 |
其他信息
- 代码库版本: v3.0
- 机器人类型: B2Z1
- 可视化: 可通过 可视化工具 查看数据集
- 引用信息: 暂无提供 BibTeX 引用信息




