据悉，群核科技聚焦于室内场景下的空间智能服务，早在2018年就发布了当时全球最大的室内空间深度 学习数据集InteriorNet，该数据集共包含1.3亿空间数据。该数据集一经发布就引发全球关注，持续被用 于AIGC、具身智能、XR等相关领域的智能训练和研究中。 近期中国AI进展不断霸榜全球，引发业内人士关注。可以看到，不论是AI模型，还是数据集，中国企 业纷纷站上全球创新前沿，这背后是中国数字经济迅猛发展的缩影，这过程中像阿里巴巴、快手、群核 科技等中国企业承载了最复杂的场景、最多元的需求，进而沉淀了最丰富的数据。而这些都将是中国企 业在AI进程中成为全球领先技术代表的基础。 本报讯 （记者梁傲男）近日，杭州群核信息技术有限公司（以下简称"群核科技"）发布的开源数据集 InteriorGS，登上全球最大AI开源社区HuggingFace趋势榜榜首，英伟达推出的Nemotron-Post数据集紧随 其后，同时上榜的还有群核开源的另一个3D数据集InteriorAgent。 当前，随着具身智能技术加速向实际应用渗透，行业正面临一个关键瓶颈：高质量训练数据的严重短 缺。AI的进化离不开数据，但机器人需要的"空间数据 ...

Group 1 - The core point of the news is that Qunhe Technology's open-source dataset InteriorGS has topped the global AI open-source community HuggingFace trend list, highlighting the increasing importance of high-quality training data in the field of embodied intelligence [1] - The InteriorGS dataset is the first to introduce 3D Gaussian technology into AI spatial training, combining realistic and semantic capabilities, making it the world's first large-scale 3D dataset suitable for intelligent agents' free movement [1] - InteriorGS consists of 1,000 Gaussian scenes, covering over 80 types of environments and more than 554,000 object labels across 755 categories, with each object accompanied by 3D bounding boxes and semantic annotations [1] Group 2 - Another dataset from Qunhe Technology, InteriorAgent, is specifically designed for the IROS 2025 "Peach Garden" and real-world robotics learning challenge, aiming to bridge the gap between simulation and reality in embodied intelligence technology [2] - The IROS competition, co-organized by Qunhe Technology and the University of Adelaide, has started registration and will hold an award ceremony on October 20 at the IROS conference [2] - Qunhe Technology has been focused on spatial intelligent services in indoor scenes, having released the world's largest indoor spatial deep learning dataset, InteriorNet, in 2018, which contains 130 million spatial data points [2]