lyl472324464/2025-11-26-twist-two-bottles-without-rinse

--- 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). ## 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": "aloha", "total_episodes": 48, "total_frames": 74418, "total_tasks": 1, "chunks_size": 1000, "data_files_size_in_mb": 200, "video_files_size_in_mb": 200, "fps": 50, "splits": { "train": "0:48" }, "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": { "observation.state": { "dtype": "float32", "shape": [ 14 ], "names": [ [ "left_waist", "left_shoulder", "left_elbow", "left_forearm_roll", "left_wrist_angle", "left_wrist_rotate", "left_gripper", "right_waist", "right_shoulder", "right_elbow", "right_forearm_roll", "right_wrist_angle", "right_wrist_rotate", "right_gripper" ] ] }, "action": { "dtype": "float32", "shape": [ 14 ], "names": [ [ "left_waist", "left_shoulder", "left_elbow", "left_forearm_roll", "left_wrist_angle", "left_wrist_rotate", "left_gripper", "right_waist", "right_shoulder", "right_elbow", "right_forearm_roll", "right_wrist_angle", "right_wrist_rotate", "right_gripper" ] ] }, "observation.velocity": { "dtype": "float32", "shape": [ 14 ], "names": [ [ "left_waist", "left_shoulder", "left_elbow", "left_forearm_roll", "left_wrist_angle", "left_wrist_rotate", "left_gripper", "right_waist", "right_shoulder", "right_elbow", "right_forearm_roll", "right_wrist_angle", "right_wrist_rotate", "right_gripper" ] ] }, "observation.effort": { "dtype": "float32", "shape": [ 14 ], "names": [ [ "left_waist", "left_shoulder", "left_elbow", "left_forearm_roll", "left_wrist_angle", "left_wrist_rotate", "left_gripper", "right_waist", "right_shoulder", "right_elbow", "right_forearm_roll", "right_wrist_angle", "right_wrist_rotate", "right_gripper" ] ] }, "observation.images.cam_high": { "dtype": "video", "shape": [ 3, 480, 640 ], "names": [ "channels", "height", "width" ], "info": { "video.height": 480, "video.width": 640, "video.codec": "av1", "video.pix_fmt": "yuv420p", "video.is_depth_map": false, "video.fps": 50, "video.channels": 3, "has_audio": false } }, "observation.images.cam_low": { "dtype": "video", "shape": [ 3, 480, 640 ], "names": [ "channels", "height", "width" ], "info": { "video.height": 480, "video.width": 640, "video.codec": "av1", "video.pix_fmt": "yuv420p", "video.is_depth_map": false, "video.fps": 50, "video.channels": 3, "has_audio": false } }, "observation.images.cam_left_wrist": { "dtype": "video", "shape": [ 3, 480, 640 ], "names": [ "channels", "height", "width" ], "info": { "video.height": 480, "video.width": 640, "video.codec": "av1", "video.pix_fmt": "yuv420p", "video.is_depth_map": false, "video.fps": 50, "video.channels": 3, "has_audio": false } }, "observation.images.cam_right_wrist": { "dtype": "video", "shape": [ 3, 480, 640 ], "names": [ "channels", "height", "width" ], "info": { "video.height": 480, "video.width": 640, "video.codec": "av1", "video.pix_fmt": "yuv420p", "video.is_depth_map": false, "video.fps": 50, "video.channels": 3, "has_audio": false } }, "task_index": { "dtype": "int64", "shape": [ 1 ], "names": null }, "subtask": { "dtype": "string", "shape": [ 1 ], "names": null }, "is_for_training": { "dtype": "bool", "shape": [ 1 ], "names": null }, "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 } } } ``` ## Citation **BibTeX:** ```bibtex [More Information Needed] ```This dataset was created using LeRobot and is designed for robotics control tasks, featuring data from an ALOHA robot with dual arms. It includes 48 training episodes, totaling 74,418 frames at a sampling rate of 50 fps. The features encompass 14-dimensional robot state observations (e.g., positions for left/right arm joints such as waist, shoulder, elbow), actions, velocities, and efforts, along with video observations from four cameras: a high-resolution camera, a low-resolution camera, a left wrist camera, and a right wrist camera, with a resolution of 480x640 and AV1 codec. The dataset is organized in parquet files, with a total data size of approximately 200 MB and video files of about 200 MB, suitable for robot learning and imitation learning research.

搜集汇总

数据集介绍

构建方式

该数据集依托LeRobot框架构建，聚焦于机器人领域的双机械臂协同操作任务，具体为无需冲洗的两瓶旋拧操作。数据采集平台采用ALOHA机器人系统，以50帧每秒的高采样率记录48个完整操作回合，累计包含74418帧原始数据。数据存储采用高效的Parquet格式分块存储（每块1000帧），同时配套存储来自四台摄像机（高视角、低视角、左右腕部视角）的同步视频流，视频编码为AV1格式，分辨率统一为640×480。数据划分为训练集，涵盖全部48个回合。

使用方法

数据集通过LeRobot库集成于HuggingFace平台，用户可直接利用该库的Dataset接口加载数据。推荐采用标准训练流程：首先，将高维视觉观测与低维状态向量对齐，构建统一的策略输入空间；其次，利用提供的‘is_for_training’标签划分训练集（当前全部可用）；最后，基于‘动作’字段与‘观测’字段训练模仿学习或强化学习模型。研究者亦可利用‘子任务’标签进行分层策略学习，或仅使用指定视角的视觉数据进行消融实验。数据以预定义索引格式组织，支持高效批处理与序列采样。

背景与挑战

背景概述

该数据集创建于2025年11月26日，由Hugging Face团队基于LeRobot框架开发，旨在为仿人机器人精细操作提供标准化训练数据。核心研究问题聚焦于双臂协调操作中的非刚性物体操控，具体任务为无冲洗条件下同时拧动两个瓶子，这要求机器人具备高精度力反馈和实时视觉感知能力。数据集采用ALOHA机器人平台，包含48个示范片段、总计74,418帧数据，以50帧/秒的速率采集了多视角视觉图像（高空、低位、左右腕相机）及14维关节状态、速度、力矩等物理量。作为机器人模仿学习领域的重要资源，该数据集填补了双机械臂协同完成复杂装配任务的数据空白，为迁移学习、示教轨迹泛化及多模态传感器融合研究提供了关键基准。

当前挑战

该数据集所解决的领域问题聚焦于仿人机器人双臂协调操作的精细控制难题，尤其在无冲洗约束下同时拧动两个瓶盖这一任务，要求机器人准确识别拧动时机、协调施加力矩并同步完成闭环动作，对力位混合控制和视觉-触觉融合范式构成严峻考验。构建过程中面临的挑战包括：高精度视频与传感器数据的同步采集，四个摄像头（640×480分辨率）与14维关节状态需在50Hz帧率下严格对齐；示范轨迹的多样性不足，48条片段可能难以覆盖真实环境的非结构化扰动；此外，AV1编码视频虽然存储高效，但解码实时性对训练流水线提出额外计算要求。数据标注的质量控制亦为难点，细微的拧动角度偏差可能影响策略泛化能力。

常用场景

经典使用场景

该数据集基于LeRobot框架构建，以ALOHA双臂机器人平台为硬件基础，专注于“扭开两个免冲洗瓶盖”这一精细操作任务。数据集中包含48个完整操作序列，共计超过74,000帧，以50Hz的高频率同步记录了机器人双臂14个关节的状态信息、动作指令、速度与力矩数据，以及来自四路视觉传感器的视频流。经典的使用场景是作为模仿学习与行为克隆算法的训练与评估基准，研究者能够利用这一高维度、多模态的时序数据，训练机器人模型学习从视觉输入到关节空间动作的端到端映射，从而赋予机器人自主执行特定精细操作的能力。

解决学术问题

在机器人学习领域，该数据集着力解决了开放环境下精细操作技能获取中数据稀缺与动作泛化性不足的学术难题。它通过提供标准化的、高保真的双臂协调操作数据，为从示教中学习（Learning from Demonstration）提供了高质量的训练样本，使得研究得以聚焦于如何从非结构化视觉信息中稳健地提取操作意图，并转化为精准的关节级控制序列。这一数据集的公开，推动了模仿学习方法在复杂接触任务中的迁移与泛化研究，对理解多指节协调、力位混合控制等前沿问题具有重要的理论支撑意义，有效促进了机器人技能学习从结构化仿真向真实世界的跨越。

实际应用

在实际应用层面，该数据集所代表的精细操作技能，其价值直接体现在对高精度、低容错场景的赋能上。例如，在医疗行业中，机器人可学习借助此类数据完成试剂瓶的自动化开盖与分液操作，从而减少人为污染风险；在食品或日化产品的柔性制造线上，双臂机器人能够像人一样拧开不同材质和扭矩的瓶盖，大幅提升产线换产的效率。此外，对于需要执行杂务的辅助机器人，掌握“拧瓶盖”这一日常技能，是其在家庭环境中实现衣物洗涤液添加、调料瓶开启等任务的必要能力基础，推动了服务机器人从可行走向可用。

数据集最近研究

最新研究方向

随着具身智能与模仿学习的迅猛发展，双臂协同操作成为机器人领域的前沿热点。该数据集基于ALOHA机器人平台，采集了48个拧瓶盖任务的高频轨迹与多视角视觉序列，为研究精细双手协作控制提供了宝贵资源。在近期研究中，该类精细操作数据被广泛用于训练基于扩散策略或Transformer架构的模仿学习模型，旨在提升机器人在非结构化环境中执行日常家务的鲁棒性与泛化能力。该数据集的发布不仅促进了机器人从简单抓取向复杂交互任务的跨越，也推动了智能家居与辅助机器人技术的实用化进程，具有重要的学术探索与产业应用价值。