Prism_vla_setupcup
收藏资源简介:
该数据集是使用LeRobot框架创建的机器人操控数据集,旨在为机器人学习任务提供标准化数据。数据集包含56个episode,共68336帧,均来自单一任务。数据以30fps的帧率采集,包含两个视角的摄像头图像(front和arm),分辨率均为480x640像素,采用AV1编码。此外,还提供了机器人关节状态(6维,包括shoulder_pan、shoulder_lift、elbow_flex、wrist_flex、wrist_roll和gripper的位置)和对应的动作指令。机器人类型为so_follower。数据集结构包括时间戳、帧索引、episode索引、任务索引等辅助字段。所有数据存储在parquet文件中,视频存储在mp4文件中。该数据集适用于机器人模仿学习、行为克隆、强化学习等任务。
This dataset is a robot manipulation dataset created using the LeRobot framework, aimed at providing standardized data for robot learning tasks. It contains 56 episodes with a total of 68,336 frames, all from a single task. The data is collected at a frame rate of 30fps, including camera images from two perspectives (front and arm), both with a resolution of 480x640 pixels, encoded using AV1. Additionally, robot joint states (6 dimensions, including positions of shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, and gripper) and corresponding action commands are provided. The robot type is so_follower. The dataset structure includes auxiliary fields such as timestamps, frame indices, episode indices, and task indices. All data is stored in parquet files, and videos are stored in mp4 files. This dataset is suitable for tasks such as robot imitation learning, behavior cloning, and reinforcement learning.
数据集概述
名称:Prism_vla_setupcup (Salim-Aissi/Prism_vla_setupcup)
任务类型:机器人学 (robotics)
许可证:Apache-2.0
创建工具:LeRobot (https://github.com/huggingface/lerobot)
数据集结构
基本信息
- 机器人类型:so_follower
- 帧率 (fps):30
- 总片段数 (episodes):56
- 总帧数 (frames):68,336
- 总任务数 (tasks):1
- 数据划分:仅训练集(片段 0 至 55)
数据内容
数据文件为 Parquet 格式,视频文件为 MP4 格式(使用 AV1 编码,分辨率 480×640,30 fps)。
- 数据文件大小:约 100 MB
- 视频文件大小:约 200 MB
特征(Features)
| 特征名 | 类型 | 维度 | 具体含义 |
|---|---|---|---|
action |
float32 | 6 | 机械臂6个关节位置(肩部旋转、肩部升降、肘部弯曲、腕部弯曲、腕部旋转、夹爪) |
observation.state |
float32 | 6 | 机械臂6个关节位置(同 action 定义) |
observation.images.front |
video | 480×640×3 | 机器人前视摄像头图像 |
observation.images.arm |
video | 480×640×3 | 机械臂视角摄像头图像 |
timestamp |
float32 | 1 | 时间戳 |
frame_index |
int64 | 1 | 帧索引 |
episode_index |
int64 | 1 | 片段索引 |
index |
int64 | 1 | 全局索引 |
task_index |
int64 | 1 | 任务索引(本数据集仅1个任务) |
数据用途
该数据集适用于机器人操控任务的模仿学习或视觉-语言-动作(VLA)模型训练,包含了从固定视角和机械臂视角拍摄的图像,以及对应的机械臂关节位置控制动作序列。
引用信息
该数据集暂无 BibTeX 引用信息(标注为 "More Information Needed")。




