Tamanenishiki/record-0430v3

Name: Tamanenishiki/record-0430v3
Creator: Tamanenishiki
Published: 2026-04-30 08:38:49
License: 暂无描述

Hugging Face2026-04-30 更新2026-05-03 收录

下载链接：

https://hf-mirror.com/datasets/Tamanenishiki/record-0430v3

下载链接

链接失效反馈

官方服务：

资源简介：

--- license: apache-2.0 task_categories: - robotics tags: - LeRobot configs: - config_name: default data_files: data/*/*.parquet --- This dataset was created using [LeRobot](https://github.com/huggingface/lerobot). <a class="flex" href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=Tamanenishiki/record-0430v3"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/> </a> ## Dataset Description - **Homepage:** [More Information Needed] - **Paper:** [More Information Needed] - **License:** apache-2.0 ## Dataset Structure [meta/info.json](meta/info.json): ```json { "codebase_version": "v3.0", "robot_type": "bi_openarm_follower", "total_episodes": 50, "total_frames": 67113, "total_tasks": 1, "chunks_size": 1000, "data_files_size_in_mb": 100, "video_files_size_in_mb": 200, "fps": 30, "splits": { "train": "0:50" }, "data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet", "video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4", "features": { "action": { "dtype": "float32", "names": [ "right_joint_1.pos", "right_joint_1.vel", "right_joint_1.torque", "right_joint_2.pos", "right_joint_2.vel", "right_joint_2.torque", "right_joint_3.pos", "right_joint_3.vel", "right_joint_3.torque", "right_joint_4.pos", "right_joint_4.vel", "right_joint_4.torque", "right_joint_5.pos", "right_joint_5.vel", "right_joint_5.torque", "right_joint_6.pos", "right_joint_6.vel", "right_joint_6.torque", "right_joint_7.pos", "right_joint_7.vel", "right_joint_7.torque", "right_gripper.pos", "right_gripper.vel", "right_gripper.torque", "left_joint_1.pos", "left_joint_1.vel", "left_joint_1.torque", "left_joint_2.pos", "left_joint_2.vel", "left_joint_2.torque", "left_joint_3.pos", "left_joint_3.vel", "left_joint_3.torque", "left_joint_4.pos", "left_joint_4.vel", "left_joint_4.torque", "left_joint_5.pos", "left_joint_5.vel", "left_joint_5.torque", "left_joint_6.pos", "left_joint_6.vel", "left_joint_6.torque", "left_joint_7.pos", "left_joint_7.vel", "left_joint_7.torque", "left_gripper.pos", "left_gripper.vel", "left_gripper.torque" ], "shape": [ 48 ] }, "observation.state": { "dtype": "float32", "names": [ "right_joint_1.pos", "right_joint_1.vel", "right_joint_1.torque", "right_joint_2.pos", "right_joint_2.vel", "right_joint_2.torque", "right_joint_3.pos", "right_joint_3.vel", "right_joint_3.torque", "right_joint_4.pos", "right_joint_4.vel", "right_joint_4.torque", "right_joint_5.pos", "right_joint_5.vel", "right_joint_5.torque", "right_joint_6.pos", "right_joint_6.vel", "right_joint_6.torque", "right_joint_7.pos", "right_joint_7.vel", "right_joint_7.torque", "right_gripper.pos", "right_gripper.vel", "right_gripper.torque", "left_joint_1.pos", "left_joint_1.vel", "left_joint_1.torque", "left_joint_2.pos", "left_joint_2.vel", "left_joint_2.torque", "left_joint_3.pos", "left_joint_3.vel", "left_joint_3.torque", "left_joint_4.pos", "left_joint_4.vel", "left_joint_4.torque", "left_joint_5.pos", "left_joint_5.vel", "left_joint_5.torque", "left_joint_6.pos", "left_joint_6.vel", "left_joint_6.torque", "left_joint_7.pos", "left_joint_7.vel", "left_joint_7.torque", "left_gripper.pos", "left_gripper.vel", "left_gripper.torque" ], "shape": [ 48 ] }, "observation.images.front": { "dtype": "video", "shape": [ 480, 640, 3 ], "names": [ "height", "width", "channels" ], "info": { "video.height": 480, "video.width": 640, "video.codec": "h264", "video.pix_fmt": "yuv420p", "video.is_depth_map": false, "video.fps": 30, "video.channels": 3, "has_audio": false } }, "timestamp": { "dtype": "float32", "shape": [ 1 ], "names": null }, "frame_index": { "dtype": "int64", "shape": [ 1 ], "names": null }, "episode_index": { "dtype": "int64", "shape": [ 1 ], "names": null }, "index": { "dtype": "int64", "shape": [ 1 ], "names": null }, "task_index": { "dtype": "int64", "shape": [ 1 ], "names": null } } } ``` ## Citation **BibTeX:** ```bibtex [More Information Needed] ```

This dataset was created by LeRobot and is primarily used in the field of robotics. It contains 50 episodes with a total of 67,113 frames at 30fps. The data is stored in parquet format, and videos are stored in mp4 format. The dataset includes multiple features such as actions (including position, velocity, and torque of left and right joints and grippers), observation states (same features as actions), front images (480x640 resolution, 3 channels), timestamps, frame indices, episode indices, etc. No detailed English description is provided for the dataset.

提供机构：

Tamanenishiki

搜集汇总

数据集介绍

构建方式

record-0430v3数据集基于LeRobot框架构建，专为机器人操作任务设计。数据采集自一款名为bi_openarm_follower的双臂机器人，该机器人每侧配备7个关节与一个夹爪，共计48维的动作与状态空间。数据以parquet格式存储，按每1000帧为一个chunk进行分块组织，便于高效加载。同时，视觉信息以h264编码的mp4视频文件保存，分辨率为640×480像素，帧率稳定在30fps。数据集共包含50个episode，总计67,113帧，均用于训练，覆盖单一操作任务。

特点

该数据集的核心特色在于其精细的多模态信息融合。每个时间步的观测数据不仅包含7+1关节组（位置、速度、扭矩）的完整状态，还同步记录对应的动作指令与前端摄像头影像，为模仿学习与行为克隆提供了丰富的特征对齐基础。数据命名规范统一，如右臂关节动作信号被清晰标注为right_joint_1.pos等，降低了特征解析的复杂度。此外，300MB的总数据体积（含100MB的parquet文件与200MB视频）在保证样本多样性的同时，兼顾了存储与传输的便利性。

使用方法

用户可通过LeRobot库直接加载该数据集。在Python环境中运行from lerobot import Dataset后，指定dataset_path为Tamanenishiki/record-0430v3即可获取数据迭代器。数据以episode为单位进行访问，每个episode包含动作序列、关节状态、视觉帧及时间戳。建议研究人员利用observation.images.front、action与observation.state三个关键字段，构建端到端的视觉-运动控制策略。数据集已完全划分为训练集（0至49号episode），无需额外进行分割处理。

背景与挑战

背景概述

record-0430v3数据集是在机器人学习领域蓬勃发展的背景下，由研究者在LeRobot框架下创建的一个面向双臂操作任务的精细数据集。该数据集以Apache-2.0许可证发布，核心研究问题聚焦于如何利用真实机器人演示数据，驱动双臂协同操作策略的学习与泛化。数据集记录了50个完整轨迹，总计超过67,000帧的高频状态与动作信息，涵盖视觉观测与多维度关节控制信号，为模仿学习与行为克隆等范式提供了坚实的训练基础。其发布推动了开放机器人数据集生态的构建，对降低机器人技能学习的数据门槛和促进可复现研究具有重要影响。

当前挑战

该数据集旨在解决的核心领域问题在于，如何从有限的专家演示中高效学习双臂机器人的精细操作能力，尤其是在高维连续动作空间与复杂动态环境下实现稳健的泛化。构建过程中面临多重挑战：其一，在仅有单一任务类型的50段演示中，需确保数据覆盖足够的状态空间，避免策略过拟合于特定场景；其二，高频采集的48维动作与状态信号（包含关节位置、速度、力矩及夹爪信息）对数据清洗、同步与存储提出了严苛要求；其三，视觉与运动数据的异构融合以及实时视频编码（H.264）的压缩保真度，进一步加剧了表征学习的难度。

常用场景

经典使用场景

在机器人操作学习领域，record-0430v3数据集为模仿学习与行为克隆算法的训练提供了高质量的数据基础。该数据集记录了双臂机器人（bi_openarm_follower）在执行特定任务时的完整运动轨迹，包含左右各七个关节的位置、速度、力矩信息以及夹爪状态，同时配套提供前视摄像头以30帧每秒采集的640×480像素视频流。研究者可将这些多模态观测量与动作序列配对，构建端到端的策略网络，使机器人能通过观测图像和关节状态复现示范操作。

衍生相关工作

该数据集催生了多项具有代表性的研究工作。基于其高保真动作与状态记录，研究者开发了适用于双臂系统的隐式行为克隆算法，通过扩散模型生成连续动作序列。此外，数据集提供的实时视频流推动了基于Transformer的视觉运动策略设计，使模型能从前置摄像头画面中提取空间-时间特征。这些衍生工作不仅验证了数据集的可用性，更为开源机器人数据集生态构建了可复现的评估标准，成为连接仿真环境与真实机器人部署的关键纽带。

数据集最近研究