pusht / README.md
smilegeng's picture
Upload README.md with huggingface_hub
3605b9f verified
|
Raw
History Blame Contribute Delete
3.17 kB
metadata
license: mit
task_categories:
  - robotics
tags:
  - LeRobot
  - tsfile
  - time-series
  - timeseries

lerobot_pusht (TsFile 格式)

本数据集由 HuggingFace 数据集 lerobot/pusht 转换为 Apache TsFile 格式。 原始数据集https://huggingface.co/datasets/lerobot/pusht

数据集简介

Push-T 是 Diffusion Policy(Chi et al. 2023)提出的机械臂操作基准任务:智能体需把一个 T 形物块推到目标位姿。本数据集使用 LeRobot 采集(codebase v2.0),是其低维状态版本——二维末端执行器状态与二维动作的时间序列。

原始数据结构

类型 说明
observation.state float32[2] 末端执行器状态(x, y)
action float32[2] 动作(x, y)
episode_index int64 episode 序号
frame_index int64 episode 内帧序号
timestamp float32 episode 内相对秒数
next.reward float32 奖励
next.done / next.success bool 终止 / 成功标志
index int64 全局样本序号
task_index int64 任务序号
observation.image video[96×96×3] 未包含,本低维版本不含图像)

转换为 TsFile 的说明

转换走「脚本预处理 + Apache TsFile Java 工具(schema 模式)」路径:

  • 数组展开observation.state[2]state_0state_1action[2]action_0action_1(保持 float32 单精度 → TsFile FLOAT)。
  • 列名清洗. 统一替换为 _next.rewardnext_reward 等)。
  • 时间轴Time = frame_index × 100ms(10 fps),时间精度 ms
  • Tag 列(设备维度)episode_idtask_id 声明为 TsFile TAG,因此每个 episode 是独立 device,各自时间轴从 0 开始、互不冲突。
  • ⚠️ 丢弃的列:原始 timestamp(episode 相对秒数,float)被丢弃,因为它在不同 episode 间会重复;改用 frame_index × 100ms 的整数毫秒时间轴。其余列全部保留。

转换后表名为 pusht,单文件 lerobot_pusht.tsfile,共 25,650 行。

用法

# 用 Apache TsFile SDK 读取 lerobot_pusht.tsfile
from tsfile import TsFileReader
reader = TsFileReader("lerobot_pusht.tsfile")
# table "pusht":tag 列 episode_id / task_id,其余为 field 测点

引用

@article{chi2024diffusionpolicy,
    author = {Cheng Chi and Zhenjia Xu and Siyuan Feng and Eric Cousineau and Yilun Du and Benjamin Burchfiel and Russ Tedrake and Shuran Song},
    title ={Diffusion Policy: Visuomotor Policy Learning via Action Diffusion},
    journal = {The International Journal of Robotics Research},
    year = {2024},
}