搬箱子、拧螺丝、上下料，第二十六届中国国际工业博览会的机器人展区，宛如一个个临时工厂，ABB机器人、发那科、安川电机等国际机器人巨头悉数设展，与智元、节卡、新松、埃斯顿、汇川技术等本土力量正面交锋，比拼的主题只有一个，谁的机器人更能干活。ABB机器人全球总裁马思康在工博会现场的这句话，正在成为行业共识，工业自动化的下一个时代，将由机器人的学习速度、而非机器人的编程水平来定义。2026年被业内视为具身智能产业交付元年，行业告别概念化探索，迈入场景化、实用化、规模化发展新阶段。
Carrying boxes, tightening screws, loading and unloading—the robot exhibition area at the 26th China International Industry Fair resembles temporary factories. International robot giants including ABB Robotics, Fanuc, and Yaskawa Electric all set up booths, confronting本土 forces including AgiBot, JAKA, Siasun, Estun, and Inovance head-on. The theme of competition is只有一个—whose robots can work better. ABB Robotics Global President Sami Atiya's statement at the fair is becoming industry consensus: the next era of industrial automation will be defined by robots' learning speed, not their programming level. 2026 is regarded by the industry as the first year of embodied intelligence industrial delivery, with the industry bidding farewell to conceptual exploration and entering a new stage of scenario-based, practical, and scaled development.
今年的工博会，ABB机器人等国际工业机器人巨头不约而同地把进厂干活作为展示重点。ABB机器人完整呈现工业级物理AI从技术构想到产业落地的全链路图景，并全球首发两款紧凑型高性能工业机器人。安川坚持工业AI务实落地路线，本次工博会展示两大技术方向，一是将AI深度嵌入自动化装备本体，新品Motoman NEXT控制柜可原生运行AI模型，落地视觉插拔、无序分拣场景，二是依托运动控制硬件底座赋能AI产业应用，新发布GP35机器人，主攻大尺寸PCB板材搬运。
At this year's Industry Fair, international industrial robot giants including ABB Robotics coincidentally made "working in factories" their display focus. ABB Robotics fully presented the full-chain picture of industrial-grade physical AI from technical conception to industrial deployment, and globally debuted two compact high-performance industrial robots. Yaskawa adheres to a pragmatic industrial AI deployment route, showcasing two major technical directions at this fair: first, deeply embedding AI into automation equipment bodies, with the new Motoman NEXT control cabinet capable of natively running AI models, deploying vision plugging and random sorting scenarios; second, leveraging motion control hardware foundations to empower AI industry applications, with the newly released GP35 robot focusing on large-size PCB handling.
本土企业的进厂步伐同样迅猛。在新松打造的人形机器人芯片封装固晶设备精密上下料工作站，上海证券报记者看到一台新松BR-MR73A全身轮式人形机器人正忙着上下料，展示人形机器人在半导体设备上下料场景中的应用能力。智元精灵G2系列机器人也是工厂里的多面手。记者在智元展台看到，精灵G2系列机器人穿梭在精密制造、柔性装配、汽车零部件上料与仓储物流等工业场景，既能干插拔光模块这种精细活，又能扛货码垛，忙得不亦乐乎。智元创新具身业务副总裁严亦行对记者表示，智元精灵G2已在客户产线上跑出可量化指标，在龙旗科技南昌工厂，精灵G2并线平板量产线精密上下料，实现场景标定最快15分钟、36小时完成集成并线、设备复用率达95%，在6天连续并线直播任务中成功率达99.99%。
Local enterprises' factory entry pace is equally rapid. At Siasun's humanoid robot chip packaging die-bonding equipment precision loading/unloading workstation, Shanghai Securities News reporters saw a Siasun BR-MR73A full-body wheeled humanoid robot busy loading and unloading, demonstrating humanoid robots' application capability in semiconductor equipment loading/unloading scenarios. AgiBot's Elf G2 series robots are also factory multi-taskers. At AgiBot's booth, reporters saw Elf G2 series robots shuttling through industrial scenarios including precision manufacturing, flexible assembly, automotive parts loading, and warehousing logistics—capable of fine tasks like plugging optical modules and heavy tasks like carrying and palletizing. AgiBot Innovation Embodied Business Vice President Yan Yixing told reporters that Elf G2 has already produced quantifiable metrics on customer production lines: at Longqi Technology's Nanchang factory, Elf G2 performs precision loading/unloading on parallel tablet mass production lines, achieving scenario calibration in as fast as 15 minutes, integration and line merging in 36 hours, and 95% equipment reuse rate, with 99.99% success rate in a 6-day continuous line-merging livestream task.
在均胜电子，精灵G2承担高难度三销定位柔性装配，节拍最快达12.97秒、成功率99.9%，体系内适配工位已达千个。节卡机器人将真实工厂复杂作业工况完整搬入展馆，直观展现产品规模化落地实力。现场工作人员介绍，公司的Kargo10轮式人形机器人今年7月已进入一家全球知名汽车零部件企业的智能大灯工厂打工，单箱作业节拍从105秒压缩至28秒，新场景部署周期缩短至57天。此外，在本届工博会上，50公斤重载升级版Kargo50全球首发。在汇川技术展台的汽车焊装作业示范线上，人形机器人联动视觉系统与弧焊机器人，完成工件上下料、复位等工序，协同开展弧焊作业，展示物理AI与机器人技术在具体生产场景中的应用探索。
At Junsheng Electronics, Elf G2 undertakes high-difficulty three-pin positioning flexible assembly, with cycle time as fast as 12.97 seconds and 99.9% success rate, with adapted workstations in the system reaching a thousand. JAKA Robotics brought real factory complex operating conditions completely into the exhibition hall, intuitively demonstrating product scaled deployment strength. On-site staff introduced that the company's Kargo10 wheeled humanoid robot entered a globally renowned auto parts enterprise's smart headlight factory in July this year, compressing single-box cycle time from 105 seconds to 28 seconds, with new scenario deployment cycles shortened to 57 days. Additionally, at this fair, the 50 kg heavy-load upgraded version Kargo50 made its global debut. At Inovance's automotive welding demonstration line, humanoid robots coordinated with vision systems and arc welding robots to complete workpiece loading/unloading, resetting, and other processes, collaboratively conducting arc welding operations, demonstrating physical AI and robot technology application exploration in specific production scenarios.
在工博会开幕式圆桌对谈环节，宇树科技创始人王兴兴表示，具身智能的ChatGPT时刻正加速到来，最快可能只需一两年。不过，热闹的进厂潮背后，具身智能距离真正稳定好用仍有不短的距离。其中，数据是横亘在规模化落地前的第一道壁垒。本届工博会上，不少企业带来了高质量数据的解决方案。例如，觅蜂科技携MEgo系列无本体数据采集终端、MEgo Engine数据治理引擎及物理AI数据众包服务平台觅蜂派亮相，展示从真实操作数据采集到数据治理的全链路技术体系及场景任务组织的服务能力，为具身智能模型训练与评测提供数据基础设施。中国工程院院士、阿里云创始人王坚在工博会上公开替机器人企业要数据。他呼吁工业界把数据放出来，让工业数据成为除语言、代码之外AI的头等数据，否则AI和工业之间的拥抱还有很大挑战。
At the Industry Fair's opening roundtable, Unitree founder Wang Xingxing said embodied intelligence's ChatGPT moment is accelerating, possibly in as little as one or two years. However, behind the bustling factory entry wave, embodied intelligence still has a considerable distance from truly stable and usable. Among them, data is the first barrier standing before scaled deployment. At this year's fair, many companies brought high-quality data solutions. For example, Mifeng Technology appeared with the MEgo series bodyless data collection terminals, MEgo Engine data governance engine, and physical AI data crowdsourcing service platform "Mifeng Pai," showcasing a full-chain technical system from real operation data collection to data governance and scenario task organization service capabilities, providing data infrastructure for embodied intelligence model training and evaluation. Chinese Academy of Engineering academician and Alibaba Cloud founder Wang Jian publicly asked for data on behalf of robot enterprises at the fair. He called on industry to release data, making industrial data AI's "first-class data" alongside language and code, otherwise the embrace between AI and industry still faces great challenges.
