广告
加载中

快递员保洁员理货员……京东发动60万人为具身智能采集数据

亿邦动力 2026-04-28 10:04
亿邦动力 2026/04/28 10:04

邦小白快读

EN
全文速览

京东发动超过60万人采集数据,推动具身智能发展,关键在于解决数据需求与采集难题。

1. 数据需求分为三层:互联网视频数据、特定场景人类第一视角实操数据、机器人本体运行数据,数据量自下而上递减,但采集难度递增。

2. 采集瓶颈包括数据不足、杂乱非标准化、难以训练和合规流通,京东通过全链路基础设施提供解决方案,如JoyEgoCam设备让员工边工作边采集。

采集规模与细节表明其可行性。

1. 计划首年完成500万小时人类活动视频采集,两年内突破1000万小时,目标成为全球最大数据服务商。

2. JoyEgoCam设备配备4K高清摄像头、130度广角等,重仅220克,便于在物流、医疗等场景使用。

处理效率高、成果显著。

1. 数据经AI数据湖清洗转换,使用JoyBuilder仿真平台泛化扩增,模型训练效率提升3.5倍,成本降低60%,数据有效率达95%。

2. 京东具身大模型JoyAI-RA真机实验成功率73.5%,上线数据交易平台支持合规流通,首批开放2000小时数据集。

京东在具身智能领域的行动提供品牌营销和产品研发启示,响应消费趋势。

1. 品牌营销通过发动60万人参与数据采集,展示社会责任和创新能力,如利用内部10万员工和外部50万人员覆盖超百场景,提升品牌形象。

2. 产品研发涉及JoyEgoCam设备的创新特性,包括4K高清、便携轻巧,支持在零售、医疗等场景应用,满足智能产品需求。

消费趋势与用户行为体现新机会。

1. 用户行为观察显示,具身智能需基于真实交互数据,如快递员、保洁员实操采集,揭示家庭、办公等场景的智能应用趋势。

2. 数据交易平台提供品牌合作渠道,支持多方协同,契合数字消费升级趋势。

京东的数据采集行动带来新增长市场和合作机会,需把握政策与风险。

1. 政策解读表明数据合规交易是趋势,京东上线具身智能数据交易平台,打通合规通道,降低流通风险。

2. 消费需求变化提供机会,如发动外部50万人参与采集,卖家可学习此模式进入物流、医疗等新兴市场。

正面影响与应对措施可借鉴。

1. 机会提示:通过JoyBuilder平台降低模型研发门槛,卖家可参与数据采集或使用高精标注数据集,首批开放2000小时。

2. 风险提示:数据杂乱非标问题已被解决,京东全链路管理提升效率95%,卖家需关注硬件标准统一以避免类似瓶颈。

京东的行动启示工厂推进数字化生产和设计需求,提供商业机会。

1. 产品生产需求涉及设备制造,如JoyEgoCam的可穿戴设计需4K摄像头、轻量化生产,可应用于物流、医疗等工业场景。

2. 商业机会包括参与数据采集服务,京东发动超10万宿迁市民,覆盖工厂环境,工厂可提供类似设备或数据支持。

数字化启示提升效率。

1. AI数据湖平台自动清洗转换数据,工厂可借鉴此技术优化生产线,实现标准数据集处理。

2. JoyBuilder仿真平台生成仿真数据,启示工厂用于产品测试,降低研发成本60%,推动智能制造。

行业发展趋势聚焦数据驱动,京东提供新技术和解决方案应对客户痛点。

1. 行业趋势显示具身智能依赖高质量数据,京东构建全链路基础设施,从采集到交易形成生态闭环。

2. 新技术包括自研JoyEgoCam设备和JoyBuilder平台,支持4K高清采集和仿真训练,解决硬件标准不一问题。

客户痛点与解决方案高效匹配。

1. 痛点如数据杂乱、非标难以训练,京东通过AI数据湖清洗对齐,数据有效率95%,成本降低60%。

2. 解决方案涉及数据交易平台,支持服务商协同开发,如处理日均数十万条数据,模型训练效率提升3.5倍。

京东的做法满足平台需求,提供招商和运营管理方案。

1. 商业需求包括数据流通和合规,京东上线具身智能数据交易平台,定向开放数据集,支持多方协同。

2. 最新做法如发动60万人采集数据,覆盖超百场景,平台商可招商外部人员参与,提升数据规模。

运营管理与风险规避关键。

1. 运营细节通过SaaS化部署可视化管理,实现视频一键上云和任务跟踪,优化流程。

2. 风险规避需关注数据孤岛,京东全链路闭环提升效率,模型成功率73.5%,可借鉴真机测试避免部署失败。

产业新动向聚焦数据瓶颈,京东商业模式和政策启示提供创新思路。

1. 产业动向显示具身智能数据采集困难,如三层数据金字塔标准不一,京东构建“硬件采集-数据处理-模型训练”闭环应对。

2. 新问题包括数据合规流通,京东上线交易平台解决孤岛现象,首批开放高精标注数据集。

商业模式与政策启示可推广。

1. 商业模式创新:从数据服务到生态共建,形成“具身智能超级供应链”,如发动10万内部员工积累100万小时机器人数据。

2. 政策建议:合规交易平台可作为法规样本,数据采集规模达500万小时,真实场景应用启示标准制定。

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

JD.com has mobilized over 600,000 people to collect data, advancing embodied intelligence by addressing data needs and collection challenges.

1. Data requirements are divided into three tiers: internet video data, first-person operational data from specific human scenarios, and robot operational data. Data volume decreases from bottom to top, but collection difficulty increases.

2. Collection bottlenecks include insufficient data, messy non-standardization, difficulty in training, and compliance in circulation. JD.com provides solutions through full-chain infrastructure, such as JoyEgoCam devices that allow employees to collect data while working.

The scale and details of data collection demonstrate its feasibility.

1. The plan aims to collect 5 million hours of human activity videos in the first year and exceed 10 million hours within two years, with the goal of becoming the world's largest data service provider.

2. The JoyEgoCam device features a 4K HD camera, 130-degree wide-angle lens, and weighs only 220 grams, making it suitable for use in logistics, healthcare, and other scenarios.

High processing efficiency and significant results.

1. Data is cleaned and transformed via an AI data lake, then generalized and augmented using the JoyBuilder simulation platform, improving model training efficiency by 3.5 times, reducing costs by 60%, and achieving a 95% data validity rate.

2. JD.com's embodied AI model, JoyAI-RA, achieved a 73.5% success rate in real-world tests. A data trading platform has been launched to support compliant circulation, with an initial release of 2,000 hours of datasets.

JD.com's initiatives in embodied intelligence offer insights for brand marketing and product development, aligning with consumer trends.

1. Brand marketing leverages the participation of 600,000 people in data collection to demonstrate social responsibility and innovation, such as utilizing 100,000 internal employees and 500,000 external participants across over 100 scenarios to enhance brand image.

2. Product development involves the innovative features of the JoyEgoCam device, including 4K HD resolution and portability, supporting applications in retail, healthcare, and other sectors to meet smart product demands.

Consumer trends and user behavior reveal new opportunities.

1. Observations of user behavior indicate that embodied intelligence requires real interaction data, such as operational data collected from couriers and cleaners, revealing trends in smart applications for home, office, and other scenarios.

2. The data trading platform provides brand collaboration channels, supporting multi-party cooperation and aligning with the trend of digital consumption upgrades.

JD.com's data collection initiative opens new growth markets and partnership opportunities, requiring attention to policies and risks.

1. Policy interpretation indicates that compliant data trading is a trend. JD.com has launched an embodied intelligence data trading platform to facilitate compliant circulation and reduce risks.

2. Changes in consumer demand present opportunities, such as involving 500,000 external participants in data collection. Sellers can adopt this model to enter emerging markets like logistics and healthcare.

Positive impacts and countermeasures can be referenced.

1. Opportunity highlights: The JoyBuilder platform lowers the barrier to model development, allowing sellers to participate in data collection or use high-precision labeled datasets, with an initial release of 2,000 hours.

2. Risk warnings: Issues with messy, non-standardized data have been resolved through JD.com's full-chain management, which improves efficiency by 95%. Sellers should focus on hardware standardization to avoid similar bottlenecks.

JD.com's initiatives provide insights for factories to advance digital production and design needs, offering commercial opportunities.

1. Product manufacturing demands involve device production, such as the wearable design of JoyEgoCam requiring 4K cameras and lightweight manufacturing, applicable in industrial scenarios like logistics and healthcare.

2. Commercial opportunities include participating in data collection services. JD.com has mobilized over 100,000 Suqian residents, covering factory environments, allowing factories to provide similar devices or data support.

Digital insights enhance efficiency.

1. The AI data lake platform automatically cleans and transforms data, enabling factories to optimize production lines and achieve standardized dataset processing.

2. The JoyBuilder simulation platform generates synthetic data, inspiring factories to use it for product testing, reducing R&D costs by 60%, and advancing smart manufacturing.

Industry trends focus on data-driven approaches, with JD.com offering new technologies and solutions to address client pain points.

1. Industry trends show that embodied intelligence relies on high-quality data. JD.com has built a full-chain infrastructure, creating an ecosystem from collection to trading.

2. New technologies include self-developed JoyEgoCam devices and the JoyBuilder platform, supporting 4K HD collection and simulation training to address hardware standardization issues.

Efficiently addressing client pain points with solutions.

1. Pain points such as messy, non-standardized data that is difficult to train are resolved through JD.com's AI data lake, achieving 95% data validity and reducing costs by 60%.

2. Solutions involve the data trading platform, enabling service providers to collaborate on development, such as processing hundreds of thousands of data entries daily and improving model training efficiency by 3.5 times.

JD.com's approach meets platform needs, offering solutions for merchant recruitment and operational management.

1. Business needs include data circulation and compliance. JD.com has launched an embodied intelligence data trading platform, selectively opening datasets to support multi-party collaboration.

2. Latest practices, such as mobilizing 600,000 people for data collection across over 100 scenarios, allow platform operators to recruit external participants to expand data scale.

Key aspects of operational management and risk avoidance.

1. Operational details are managed through SaaS-based deployment, enabling video uploads to the cloud and task tracking for process optimization.

2. Risk avoidance requires addressing data silos. JD.com's full-chain closed-loop improves efficiency, with a 73.5% model success rate, offering insights from real-world testing to avoid deployment failures.

Industry trends highlight data bottlenecks, with JD.com's business model and policy insights offering innovative ideas.

1. Industry trends indicate challenges in embodied intelligence data collection, such as inconsistencies in the three-tier data pyramid. JD.com addresses this with a 'hardware collection-data processing-model training' closed loop.

2. New issues include compliant data circulation. JD.com's trading platform resolves data silos, with an initial release of high-precision labeled datasets.

Business model and policy insights can be widely applied.

1. Business model innovation: Evolving from data services to ecosystem co-creation, forming an 'embodied intelligence super supply chain,' such as leveraging 100,000 internal employees to accumulate 1 million hours of robot data.

2. Policy recommendations: The compliant trading platform can serve as a regulatory model, with data collection reaching 5 million hours, providing real-world insights for standard setting.

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.

【亿邦原创】今年,具身智能赛道的融资和技术都指向同一个关键词——“数据”。简单地说,要让机器人学会打扫卫生整理房间,必须有关于打扫卫生整理房间的数据,让机器人去学习。所以,数据驱动,是具身智能通往GTP时刻的核心路径。

那么,具身智能需要什么样的数据?

大致可分为三层:第一层是互联网上的各种视频,尤其是人类第一视角开放场景的数据;第二层是特定场景、任务下人类第一视角实操作业数据;第三层是机器人本体自主或者由人类遥操执行任务的数据。三层数据金字塔,自下而上数据量越来越小,同时采集的难度和成本也越来越大。

当下,数据采集已成为制约具身智能发展的瓶颈:高质量、真实交互数据的不足,硬件标准不一、数据采集流程分散、标注与训练环节割裂,数据孤岛现象突出,采集到的数据杂乱、非标,难以直接用于模型训练,更无法合规流通交易。

针对这一难题,京东推出全链路具身智能数据基础设施,打通从数据采集到模型测试的全流程闭环,将原始杂乱数据精炼为驱动模型进化的高价值“数据燃料”。

数据采集环节,京东云自研的可穿戴式超高清采集终端JoyEgoCam,可在物流、零售、医疗、家庭等多种场景下“即戴即采”,让快递员、保洁员、理货员一边工作一边完成专业级数据采集。

据悉,在清晰度方面,JoyEgoCam配备4K高清摄像头,支持60帧帧率与130度超广角拍摄,可实现毫秒级动作细节捕捉,精准记录各类场景下的细微操作;在精准度方面,重投影误差小于0.2像素,搭配京东云自研立体校正技术,能够真实还原操作现场的空间立体感;在便携性方面,整机仅重220克,轻于普通智能手机,佩戴舒适;在稳定性方面,内置车规级6轴IMU与多传感器融合单元,即使在极端抖动场景下也能稳定追踪拍摄。

谁来采集数据?京东利用其丰富业务场景的优势,发动内部超过10万名各类职业员工,以及外部最多50万各行业人员,其中在宿迁就将发动超10万市民参与,覆盖家庭、办公室、工厂到物流、商店、餐厅、医疗、环卫等超百个细分场景,遍布人类真实活动的方方面面,开展“人类历史上规模最大的数据采集行动”。

根据规划,京东将在首年完成500万小时人类活动视频数据采集,两年内突破1000万小时,同时积累100万小时机器人本体运行数据。这一数据规模将使其成为全球最大的具身智能数据服务商。

数据经采集后,进入上传和加工阶段。京东云通过任务、人员、设备全流程可视化管理与SaaS化部署,实现视频一键上云。数据汇入AI数据湖平台后,自动完成清洗、对齐、转换与预标注,转为标准训练集。JoyBuilder仿真平台批量生成高逼真仿真数据,从而实现人类操作数据→仿真操作数据→真机操作数据的高效数据增值转换与泛化扩增。

治理后的数据汇聚至JoyBuilder模型开发平台,数据“开箱即训”、模型“一键部署”,模型训练效率提升3.5倍,大幅降低VLA大模型研发门槛。其自研AI算子矩阵贯穿始终,涵盖去畸变、语义描述、深度重建等关键环节,精炼高价值训练素材。目前,京东日处理数据量已达数十万条,数据有效率高达95%,整体处理成本降低60%。

可见,依托全链路基础设施,京东构建起“数据采集-模型训练-数据优化”的生态闭环。以自采数据为核心训练的京东具身大模型JoyAI-RA,在真机实验上成功率达到73.5%,超过pi0.5等SOTA模型。

此外,京东上线了具身智能数据交易平台,汇聚京东丰富业务场景下的多模态数据资源,支持数据方、开发者、应用方多方协同,打通具身智能数据合规交易通道。平台首批定向开放2000小时高精标注数据集。

可见,京东正构建起“硬件采集-数据处理-模型训练-仿真测试-合规交易-生态共建”的完整产业链条,推动具身智能从实验室研发迈向规模化商业落地,形成“具身智能超级供应链”。


文章来源:亿邦动力

广告
微信
朋友圈

这么好看,分享一下?

朋友圈 分享

APP内打开

+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