广告
加载中

诺亦腾机器人在WRC发布HiPHI 开源617.5小时高精度人体运动数据

龚作仁 2026-08-22 07:40
龚作仁 2026/08/22 07:40

邦小白快读

EN
全文速览

本文核心内容是诺亦腾机器人在2026年世界机器人大会发布开源高精度人体运动数据集HiPHI,相关干货信息总结如下:

1. 数据集核心信息:该数据集面向人形机器人研究、数字人开发领域开放,总采集时长617.5小时,包含371.8小时全身人体运动数据、245.7小时同步记录物体状态的人-物交互数据,由132名动捕演员以亚毫米级光学精度采集,按动作语义结构化整理,目前已在Hugging Face公开上线。

2. 价值与落地情况:该数据集解决了行业长期存在的数据规模与精度难以兼得的痛点,是目前公开规模最大的高精度人体运动数据集之一,基于其训练的模型已在宇树G1人形真机实现跑步、搬运等多种动作,科研可免费使用,商业使用可对接品牌申请授权。

本次诺亦腾发布HiPHI数据集的操作,给人形机器人、具身智能领域品牌商提供了多维度参考,干货总结如下:

1. 品牌营销与渠道建设参考:诺亦腾选择将核心资产高质量数据公开,依托WRC、RO-MAN两个顶级行业大会发布传播,既向行业展示了自身年产能超10万小时的强大数据生产能力,也建立了行业认可度,这种开放生态营销的思路值得同领域品牌借鉴。

2. 消费与产业趋势观察:当前Physical AI发展的核心瓶颈是高质量可用数据稀缺,同时具备大规模、高精度、人-物同步交互的数据集缺口极大,产业端和终端对落地人形机器人的需求持续上涨,数据生产赛道有明确的增长空间。

3. 产品研发方向:产品研发需要围绕动作覆盖度、逐帧精度、交互状态同步三个核心需求打磨,才能满足市场要求。

本次HiPHI数据集的发布,给人形机器人、AI数据领域相关卖家释放了明确的行业信号与机会,干货总结如下:

1. 行业机会判断:当前具身智能、人形机器人正处于高速发展阶段,高质量结构化运动数据是全行业的核心刚需,长期存在稀缺性,围绕高精度数据采集、整理、生产的相关赛道有大量增长机会,新卖家也可以切入细分赛道寻找生存空间。

2. 商业模式可参考:诺亦腾采用科研免费开放、商业收取授权费的模式,通过免费数据集培育行业生态,进而转化商业客户,这种轻推广重生态的模式适合数据赛道卖家学习。

3. 风险提示:数据生产行业存在规模和精度难以兼得的痛点,对技术能力和生产成本控制要求较高,核心技术不足的卖家很难在赛道中存活,进入前需要评估自身技术实力。

本次HiPHI数据集的发布,给布局人形机器人相关业务的工厂带来了诸多启示,干货总结如下:

1. 产品生产与设计需求:当前人形机器人训练对高精度人体运动数据、同步人-物交互数据的需求极强,工厂为人形机器人企业提供硬件产品时,需要匹配数据层的精度要求,产品设计要考虑真机训练动作迁移的适配性,才能满足客户要求。

2. 数字化转型启示:诺亦腾已经搭建了成熟的标准化数据生产基础设施,可实现年产能10万小时高质量数据,证明规模化、标准化的数据生产已经可以落地,工厂布局相关业务可以参考这套标准化生产思路,通过规模化生产降低单位成本,提升动作覆盖范围。

3. 商业机会挖掘:当前市场中高质量公开数据稀缺,工厂可以依托自身硬件产能优势,和诺亦腾这类专业数据企业合作,共同开发数据产品,拓展新的营收渠道,寻找新的增长点。

本次发布透露出具身智能行业的核心痛点与发展方向,给数据服务、AI服务领域服务商提供了明确的方向,干货总结如下:

1. 行业核心客户痛点:当前Physical AI发展的核心瓶颈不是数据总量不足,而是真正能被机器学习的高质量数据不足,行业长期存在数据规模与精度难以兼得的问题,高质量数据大多被企业作为核心资产不开放,客户缺少可信赖的公共训练数据基础,这就是服务商的机会。

2. 行业发展趋势:未来全模态、大规模、高精度的结构化数据会成为具身智能行业的核心需求,开放数据集会成为推动行业发展的重要基础,诺亦腾也计划后续发布更多全模态交互数据,数据服务赛道的增长空间十分广阔。

3. 解决方案参考:服务商可以参考诺亦腾的思路,从数据设计阶段就围绕动作覆盖度、逐帧数据质量、交互物体状态同步三个核心要求搭建生产体系,满足客户从科研到商业落地的全场景需求。

本次HiPHI数据集在Hugging Face上线开放,给AI相关平台商带来了很多可参考的经验与方向,干货总结如下:

1. 行业需求洞察:当前科研机构和AI企业对高质量公开数据集的需求极强,平台引入这类优质公开数据资源,可以有效丰富平台生态,提升平台对开发者和企业用户的吸引力,增强平台的核心竞争力。

2. 运营管理优化方向:平台可以针对数据集产品设置专门的分类管理和授权对接模块,满足科研用户免费使用、商业用户对接授权的不同需求,优化不同类型用户的使用体验。

3. 机会与风险提示:具身智能是AI领域接下来的核心增长方向,围绕数据服务的生态会快速发展,平台可以提前布局,推出针对数据生产企业的招商与扶持政策,吸引优质企业入驻抢占先机;同时需要注意数据版权与采集合规问题,规避相关运营风险。

本次诺亦腾发布HiPHI数据集,给具身智能领域研究者带来了新的资源与产业动向参考,干货总结如下:

1. 产业新动向:当前Physical AI产业已经形成共识,高质量可用数据是行业发展的核心瓶颈,头部商业数据企业已经开始开放核心数据集培育行业生态,产业已经进入数据产能竞争阶段,诺亦腾可实现年产能10万小时高质量数据,证明规模化生产高精度数据已经落地,产业进入新的发展阶段。

2. 研究资源利好:HiPHI是目前公开发布规模最大的高精度人体运动数据集之一,具备行业领先的动作覆盖度和最低的逐帧物理误差,免费对科研社区开放,可用于人形机器人动作学习、数字人运动模型研究等多个方向,还提供完整的论文和文档支撑,解决了过往研究中高质量数据不足的痛点。

3. 商业模式参考:诺亦腾采用科研免费开放、商业授权变现的模式,持续推出更多开放数据集培育生态,这种模式兼顾了行业影响力建立和商业变现,值得深入研究。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

This article centers on Noitom Robotics' launch of its open-source high-precision human motion dataset HiPHI at the 2026 World Robot Conference. Key takeaways are as follows:

1. Core dataset details: Open to researchers in humanoid robotics and digital human development, the dataset totals 617.5 hours of capture time, including 371.8 hours of full-body human motion data and 245.7 hours of synchronized human-object interaction data. Captured from 132 motion capture actors with sub-millimeter optical precision, it is structured by action semantics and is now publicly available on Hugging Face.

2. Value and access: HiPHI addresses the longstanding industry pain point of the trade-off between dataset size and precision. It ranks among the largest high-precision open human motion datasets available today. Models trained on HiPHI have already enabled capabilities such as running and object carrying on the Unitree G1 humanoid robot. Free access is provided for academic research, while commercial users can contact Noitom to apply for licensing.

Noitom's launch of the HiPHI dataset offers multi-dimensional insights for brands operating in humanoid robotics and embodied intelligence. Key takeaways are as follows:

1. Reference for brand marketing and channel building: Noitom has opened access to its high-quality core data and launched the dataset via two top industry events, the World Robot Conference and RO-MAN. This strategy showcases Noitom's strong data production capacity of over 100,000 hours of annual output, while building widespread industry recognition. This open ecosystem marketing approach is a valuable reference for peers in the sector.

2. Observation of consumer and industry trends: The core bottleneck for current Physical AI development is the scarcity of usable high-quality data. There is a huge gap for datasets that combine large scale, high precision, and synchronized human-object interaction. As demand for commercial humanoid robots continues to rise from both industry and end users, the data production track has clear growth potential.

3. Product R&D direction: To meet market demand, product development must focus on three core requirements: action coverage, per-frame precision, and synchronized interaction state recording.

The launch of the HiPHI dataset sends clear industry signals and reveals opportunities for sellers active in humanoid robotics and AI data sectors. Key takeaways are as follows:

1. Industry opportunity assessment: Embodied intelligence and humanoid robotics are currently in a period of rapid growth, and high-quality structured motion data is a core industry-wide need that will remain scarce long term. Segments related to high-precision data collection, organization and production offer substantial growth opportunities, and even new entrants can find space by focusing on niche segments.

2. Reference business model: Noitom uses a model of free access for research and licensing fees for commercial use, nurturing the industry ecosystem through free datasets to convert commercial customers. This 'ecosystem over heavy promotion' model is well suited for data track sellers to learn from.

3. Risk warning: The data production industry faces the inherent challenge of balancing scale and precision, placing high demands on technical capability and production cost control. Sellers without core technical advantages are unlikely to survive in this market, so a thorough assessment of technical strength is required before entering.

The launch of the HiPHI dataset offers multiple insights for factories developing humanoid robotics-related businesses. Key takeaways are as follows:

1. Product manufacturing and design requirements: Humanoid robot training today has extremely high demand for high-precision human motion data and synchronized human-object interaction data. When supplying hardware to humanoid robotics companies, factories need to align with data layer precision requirements, and design products to support action transfer for real-machine training to meet client needs.

2. Insights for digital transformation: Noitom has built mature, standardized data production infrastructure capable of outputting 100,000 hours of high-quality data annually, proving that large-scale, standardized data production is commercially viable. Factories entering this space can adopt this standardized production approach to reduce unit costs and expand action coverage through scaling.

3. Commercial opportunity挖掘: High-quality open data is currently scarce in the market. Factories can leverage their existing hardware production advantages to partner with specialized data companies like Noitom to co-develop data products, expand new revenue streams, and unlock new growth points.

The launch reveals the core pain points and development direction of the embodied intelligence industry, offering clear guidance for data and AI service providers. Key takeaways are as follows:

1. Core client pain points: The core bottleneck for current Physical AI development is not insufficient total data volume, but a lack of high-quality data that can actually be used for machine learning. The trade-off between dataset size and precision has long plagued the industry, and most high-quality data is held as private core assets by companies, leaving clients without access to reliable public training data resources — this gap is the key opportunity for service providers.

2. Industry development trends: Full-modal, large-scale, high-precision structured data will become the core demand of the embodied intelligence industry in the future, and open datasets will be a critical foundation for industry development. Noitom also plans to release more full-modal interaction data in the future, indicating substantial growth potential for the data service sector.

3. Reference solution framework: Service providers can follow Noitom's example by building production systems around three core requirements starting from the data design stage: action coverage, per-frame data quality, and synchronized interaction object state tracking, to meet client needs across all scenarios from academic research to commercial deployment.

The public listing of the HiPHI dataset on Hugging Face offers valuable experience and direction for AI-related platform operators. Key takeaways are as follows:

1. Industry demand insight: Research institutions and AI companies currently have extremely strong demand for high-quality open datasets. Introducing these high-quality open data resources can effectively enrich platform ecosystems, improve platform appeal to developers and enterprise users, and strengthen core platform competitiveness.

2. Operations and management optimization: Platforms can create dedicated category management and licensing对接 modules for dataset products, to accommodate the different needs of academic users seeking free access and commercial users requiring licensing, and improve the user experience for all user groups.

3. Opportunity and risk warning: Embodied intelligence is the core growth direction of the AI sector in the coming period, and the data service ecosystem will expand rapidly. Platforms can prepare in advance by launching investment attraction and support policies for data production companies to attract high-quality players and seize first-mover advantage; at the same time, platforms must pay close attention to data copyright and collection compliance to avoid operational risks.

Noitom's launch of the HiPHI dataset provides embodied intelligence researchers with new resources and insights into industry trends. Key takeaways are as follows:

1. New industry trends: The Physical AI industry has now reached a consensus that high-quality usable data is the core bottleneck for industry development. Leading commercial data companies have begun opening their core datasets to nurture the industry ecosystem, and the sector has entered an era of competition around data production capacity. Noitom's annual output of 100,000 hours of high-quality data proves that large-scale production of high-precision data is commercially viable, marking the entry of the industry into a new development stage.

2. Benefits for research resources: HiPHI is one of the largest high-precision open human motion datasets ever published, with industry-leading action coverage and the lowest per-frame physical error. It is free for the research community, can be used for research into humanoid robot action learning, digital human motion modeling and other directions, and is supported by complete papers and documentation. It addresses the longstanding problem of insufficient high-quality data for academic research.

3. Reference business model: Noitom operates on a model of free open access for research and commercial licensing for monetization, and continues to release new open datasets to nurture the ecosystem. This model balances the building of industry influence and commercial monetization, making it worthy of in-depth study.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

现已上线 Hugging Face 平台,成为诺亦腾机器人基于“World Compiler”理念公开发布的首个数据集

北京,2026年8月19日——诺亦腾机器人(Noitom Robotics,简称NR)今日在北京世界机器人大会(WRC)正式发布HiPHI(读音同“hi-fi”)。这是一套面向人形机器人学习、数字人与计算机图形学领域研究人员及工程师的高精度光学动作捕捉数据集。该数据集现已在Hugging Face公开上线,也是诺亦腾机器人基于“World Compiler”理念发布的首个公开数据集,旨在让真实物理世界成为人工智能系统可学习的对象。

此次发布的数据总时长达到617.5小时,其中包含左右镜像数据:371.8小时的全身人体运动数据,以及245.7小时的人—物交互数据。在人—物交互数据中,每件物体的运动轨迹与三维网格体(3D Mesh)均与动捕演员的动作同步记录。数据来自132名动捕演员,以90 Hz频率、亚毫米级光学动捕精度完成采集,并依据FrameNet对人类动作语义的组织框架,以动作单元进行结构化整理。

HiPHI同时提供完整的数据文档,包括由诺亦腾机器人研究团队撰写的学术论文。论文显示,在参与对比的数据集中,HiPHI具有经量化评估的最广泛动作覆盖度,以及最低的逐帧物理质量误差。基于HiPHI训练的模型已经在宇树G1人形机器人真机部署中实现跑步、坐下、爬行、搬运箱子和拖拉行李箱等动作。

“Physical AI面临的真正瓶颈,不在于世界上存在多少数据,而在于其中究竟有多少能够真正被机器学习。”诺亦腾机器人创始人兼首席执行官戴若犁博士表示,“HiPHI是我们给出的第一个答案:在采集第一帧数据之前,我们就完成了覆盖度设计;数据精度能够满足从训练到真实机器人的迁移;人与物体发生接触和交互时,物体本身也被作为运动的一部分同步记录。“这类高质量数据通常被企业视为核心资产,很少对外开放。我们选择公开HiPHI,一方面是希望为具身智能行业提供一套可信赖的公共数据基础;另一方面,也源于我们对自身持续生产高质量数据能力的信心。”

高质量人体运动数据长期稀缺,行业一直面临规模与精度难以兼得的问题:互联网视频体量庞大,却缺少机器人学习所需的精确物理状态和交互信息;实验室采集的数据精度更高,但受制于采集成本和生产效率,往往只能覆盖有限的行为类型。正因如此,同时具备大规模、高精度和丰富动作覆盖的数据集很少对外开放,由商业数据生产企业完整公开的案例更为罕见。根据公开资料,HiPHI是迄今公开发布的规模最大的高精度人体运动数据集之一。

对于研究人员,HiPHI可用于人形机器人动作跟踪、模仿学习、动作生成,也能够为数字人和计算机图形学中的运动先验模型提供高质量数据基础。对于Physical AI企业,此次公开的数据集则提供了一个可直接评估的标准样本,帮助其完整深入地了解诺亦腾机器人面向商业合作伙伴持续生产的数据类型与质量水平。

“动作覆盖度、逐帧数据质量和交互物体状态,直接决定了人形机器人能否真正从人类运动数据中学到能力。因此,从数据设计、生产评估到发布,我们始终围绕这三个重点展开。”诺亦腾机器人首席科学家韩磊博士表示,“此次发布已经包含研究人员使用和评估数据所需的完整内容,包括标准化BVH文件、同步物体轨迹、语义化动作索引。HiPHI也是我们日常数据生产能力的真实体现:依托这套基础设施,我们每年可为合作伙伴生产超过10万小时的高质量数据。”

HiPHI现已在Hugging Face上线(huggingface.co/datasets/noitomrobotics/HiPHI),并根据ModalityNet开放研究许可协议(ModalityNet Open Research License)向科研社区免费开放;如需商业授权,可通过modalitynet.com与诺亦腾机器人联系。

2026年8月24日至28日,诺亦腾机器人还将在日本福冈举行的RO-MAN 2026大会上继续开展相关发布活动。公司未来将继续开放更多数据集,并计划于今年晚些时候发布一套全模态交互数据语料。后续版本还将陆续支持SMPL、SOMA等主流数据格式。

关于诺亦腾机器人

诺亦腾机器人是一家专注于Physical AI全模态数据的数据公司。通过ModalityNet(modalitynet.com)——其“World Compiler”理念的具体实现载体,诺亦腾机器人致力于将真实世界中的人体运动、交互和行为转化为机器可学习的结构化表征。围绕这一平台,公司构建了三大数据语料体系:HiPHI-MOV(人体运动)、HiPHI-OM(全模态交互)和ITW(真实世界场景),为具身智能模型提供高质量、结构化且可直接用于训练的数据基础。

注:文/龚作仁,文章来源:Laborer,本文为作者独立观点,不代表亿邦动力立场。

文章来源:Laborer

广告
微信
朋友圈

FAQ回顾

HiPHI是什么?

HiPHI是诺亦腾机器人2026年于世界机器人大会发布的高精度光学动作捕捉数据集,总时长617.5小时,包含371.8小时全身人体运动数据、245.7小时人-物交互数据,亚毫米级采集精度,现已在Hugging Face面向科研社区免费开源。

HiPHI数据集有哪些核心优势?

HiPHI是目前公开发布的规模最大的高精度人体运动数据集之一,拥有经量化评估的最广泛动作覆盖度、最低的逐帧物理质量误差,同步记录人-物交互的物体运动轨迹与三维网格体,按动作单元结构化整理。

HiPHI数据集可应用在哪些领域?

HiPHI可用于人形机器人动作跟踪、模仿学习、动作生成,也能为数字人和计算机图形学的运动先验模型提供高质量数据基础,还可作为Physical AI企业评估数据质量的标准样本。

诺亦腾机器人的核心业务是什么?

诺亦腾机器人是专注于Physical AI全模态数据的企业,依托ModalityNet平台构建三大数据语料体系,每年可生产超10万小时高质量数据,可为具身智能模型提供可直接用于训练的结构化高质量数据。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0