具身智能通用场景多模态交互高质量数据集

具身智能通用场景多模态交互高质量数据集数据来源主要为1. 合作采集：与ZY、LC等头部机器人厂商联合采集，获取办公、康养、居家、物流等场景具身智能原始数据，含官方授权有效标注数据；2. 自主采集：基于ZY精灵 G1 本体、睿研灵巧手、遥操手套等设备，自主采集商超、奶茶店、会议室、实验室等高价值场景多模态交互数据,合规整合通用场景具身智能数据资源。数据集范围涵盖工业、商贸、居家、康养、物流等多场景机器人交互。数据总量1500小时，数据大小1PB，含 Agibot、rrd 格式原始数据及 JSON 标注文件。产出的高质量数据集主要用于模型训练和申报数据产权。

High-Quality Multimodal Interaction Dataset for General-Scene Embodied Intelligence
The data sources of this dataset mainly include two parts:
1. Cooperative collection: Jointly collected with leading robot manufacturers such as ZY and LC, original embodied intelligence data from scenarios including office, elderly care, home, logistics, etc., with officially authorized and valid annotated data;
2. Independent collection: Based on devices such as ZY Genie G1 robot body, Ruiyan Dexterous Hand, and teleoperation gloves, independently collected multimodal interaction data from high-value scenarios including supermarkets, milk tea shops, conference rooms, and laboratories, and legally integrated general-scene embodied intelligence data resources.

The dataset covers robot interactions in multiple scenarios including industrial, commercial, home, elderly care, and logistics. The total data volume is 1500 hours, with a total size of 1 PB, including original data in Agibot and rrd formats as well as JSON annotation files. This high-quality dataset is mainly used for model training and data property right declaration.

搜集汇总

背景与挑战

背景概述

该数据集通过合作与自主采集方式，整合了办公、康养、居家、物流、商超等多场景下的具身智能多模态交互数据，总量达1500小时（1PB），包含原始数据和JSON标注文件。它主要用于支持模型训练和数据产权申报。