在位于佛山顺德的美的全球创新中心，一台六臂人形机器人正在进行打螺丝测试。据悉，这款机器人打螺丝的效率已经基本赶上一位熟练产业工人，而它身上不少组件，都来自大湾区一小时配套圈。10月21日，南方网、粤学习记者走访美的库卡机器人、极亚精机、天太机器人、华沿机器人等多家大湾区具身智能机器人产业链上下游企业。从核心传动零部件研发，到整机本体集成、场景落地验证，一条分工清晰、紧密协同的机器人产业链在这里展开。
At Midea's Global Innovation Center in Shunde, Foshan, a six-arm humanoid robot is undergoing screw-driving tests. Reportedly, this robot's screw-driving efficiency has basically caught up with a skilled industrial worker, and many of its components come from the Greater Bay Area's one-hour supporting circle. On October 21, Southern Network and Yuexue reporters visited Midea KUKA Robotics, Jiya Precision Machinery, Tiantai Robotics, Huayan Robotics, and other upstream and downstream enterprises in the Greater Bay Area embodied intelligence robot supply chain. From core transmission component R&D to complete machine integration and scenario deployment validation, a clearly divided, tightly collaborative robot supply chain unfolds here.
在大湾区，人形机器人的研发迭代有多快。天太机器人运营总监李新鑫给出了答案，3到5个月。从图纸到真正的机器人样机下线，3到5个月足矣。天太机器人的上下游企业70%以上都在大湾区，沟通成本更低、响应效率更高。李新鑫表示，大湾区的具身智能机器人全产业链优势在全国具有明显优势，天太机器人的所有零部件和所用的软硬件技术都可以在大湾区的产业链中获取。华沿机器人首席发展官兼董事局主席助理林深表示，大湾区对人形机器人研发和样机迭代最独特的价值，是供应链的密度、响应速度和工程协同深度。在大湾区，从出图纸到打样到测试，全链条都能在区域内找到配合的上下游，部分工件能做到早上发出打样需求、晚上拿到样品。
How fast is humanoid robot R&D iteration in the Greater Bay Area. Tiantai Robotics Operations Director Li Xinxin gave the answer: 3 to 5 months. From drawings to a real robot prototype rolling off, 3 to 5 months is enough. Over 70% of Tiantai Robotics' upstream and downstream enterprises are in the Greater Bay Area, with lower communication costs and higher response efficiency. Li Xinxin said the Greater Bay Area's full embodied intelligence robot supply chain advantage is nationally obvious; all of Tiantai Robotics' components and software-hardware technologies can be obtained within the Greater Bay Area supply chain. Huayan Robotics Chief Development Officer and Board Chairman Assistant Lin Shen said the Greater Bay Area's most unique value for humanoid robot R&D and prototype iteration is supply chain density, response speed, and engineering collaboration depth. In the Greater Bay Area, from drawings to proofing to testing, the entire chain can find cooperating upstream and downstream within the region; some workpieces can go from "proofing request sent in the morning to sample received at night."
从产业链协同性来看，广东已构建起上下游高效联动、跨区域跨领域深度融合的协同发展生态。数据显示，依托一小时供应链和全球最完善的电子信息产业链，珠三角可实现48小时快速打样、7天小批量交付，零部件采购本地化率突破60%，整体国产化率超90%。摩根士丹利发布的2026年全球人形机器人产业链百强榜单，广东有11家企业入选，具身智能机器人产业生态集聚效应显著。为什么物理距离如此重要。人形机器人是软硬一体的复杂系统，改一个关节，往往连带改结构、改装配、改测试，迭代速度靠的是打样厂、装配线、测试台之间实实在在的距离。距离越短，改错成本越低，迭代速度越快。
On supply chain collaboration, Guangdong has built an ecosystem of efficient upstream-downstream linkage and deep cross-region, cross-field integration. Data shows relying on the one-hour supply chain and the world's most complete electronic information industry chain, the Pearl River Delta can achieve 48-hour rapid proofing and 7-day small-batch delivery, with component procurement localization exceeding 60% and overall domestic content exceeding 90%. In Morgan Stanley's 2026 Global Humanoid Robot Supply Chain Top 100 list, Guangdong has 11 enterprises selected, with significant embodied intelligence robot industry ecosystem clustering effects. Why is physical distance so important. Humanoid robots are complex software-hardware integrated systems; changing one joint often entails changing structure, assembly, and testing. Iteration speed depends on the real distance between proofing factories, assembly lines, and test benches. The shorter the distance, the lower the error-correction cost, the faster the iteration.
快，只是故事的一半。图纸变样机容易，样机要变成产线上靠得住的机器，是另一道坎。库卡具身机器人创新中心主任奚伟表示，人形机器人有柔性化能力，但到了工厂一线，在搬箱、打螺丝等工作上，它的效率低，只有人的30%。库卡的解法是六臂，靠单位面积里更多的手臂把效率堆出来。奚伟介绍，库卡的六臂人形机器人在产线上可以实现10秒打6颗螺丝，效率基本接近熟练的产业工人。华沿的账，算在力上，即精密装配和柔性插拔的力控确定性。在光模块、精密连接器等场景，碰撞风险极高，插偏一点就损伤接口。华沿七轴人形手臂搭载全关节高精度力控与阻抗控制，末端接触力精度可达0.1N。
Fast is only half the story. Drawings to prototypes is easy; turning prototypes into reliable machines on production lines is another hurdle. KUKA Embodied Robot Innovation Center Director Xi Wei said humanoid robots have flexibility, but on factory floors, in tasks like box handling and screw driving, their efficiency is low—only 30% of humans. KUKA's solution is six arms—stacking efficiency through more arms per unit area. Xi Wei introduced that KUKA's six-arm humanoid robot can drive 6 screws in 10 seconds on the line, efficiency basically approaching skilled industrial workers. Huayan's account is calculated on force—the force-control certainty of precision assembly and flexible insertion. In optical modules, precision connectors, and similar scenarios, collision risk is extremely high; slight misalignment damages interfaces. Huayan's seven-axis humanoid arm carries full-joint high-precision force control and impedance control, with end contact force precision reaching 0.1N.
但真正的难题，是仿真与现场之间的落差。对于机器人而言，仿真环境里效果很好的方案，一旦进入真实车间，面对光照变化、来料偏差、振动、油污、节拍压力，任务成功率往往会明显下滑。林深说，实验室不会告诉你连续运行后的温升、来料姿态的变化会怎么影响任务成功率，这些只有真产线会告诉你。库卡机器人已经在美的工厂部署，华沿机器人在船舶工厂做焊接。广东正在探索实景式训练、小批量应用、规模化量产三阶法推进路径。目前，全省已有超2000台人形机器人进入制造企业，在装配、搬运等工序稳定落地应用，全年部署数量有望突破3000台。
But the real difficulty is the gap between simulation and the site. For robots, solutions that work well in simulation often see明显 success rate declines once entering real workshops facing lighting changes, incoming material deviations, vibration, oil stains, and cycle time pressure. Lin Shen said: "The laboratory won't tell you how temperature rise after continuous operation or changes in incoming material posture affect task success—only real production lines tell you that." KUKA robots are already deployed in Midea factories; Huayan robots do welding in shipbuilding factories. Guangdong is exploring a three-stage path of "real-scenario training—small-batch application—scaled mass production." Currently over 2,000 humanoid robots have entered manufacturing enterprises province-wide, stably deployed in assembly, handling, and other processes, with full-year deployment expected to break 3,000 units.
要让机器人的账算得过来，整机之外还有绕不开的一环，核心零部件。以重载机器人最核心的零部件RV减速器为例，其长期被国外品牌卡脖子。在极亚精机，记者见到了机器人产业链的地基。这家美的集团旗下全资子公司，专注RV减速器、重载伺服电机和关节模组。极亚精机人形关节模组业务负责人许培林介绍，以往国产减速器跑了一段时间测试就磨齿，精度越用越差。而目前，极亚精机的RV减速器在精度、精度保持性、可靠性、寿命等维度，均已达到国际领先水平。支撑它的，是佛山、上海以及日本大阪三大研发中心，以及美的30多年的电机技术沉淀。
For robots' accounts to work out, beyond complete machines there is an unavoidable link—core components. Taking RV reducers, the most core component of heavy-load robots, as an example, they have long been bottlenecked by foreign brands. At Jiya Precision Machinery, reporters saw the robot supply chain's foundation. This wholly-owned Midea Group subsidiary focuses on RV reducers, heavy-load servo motors, and joint modules. Jiya Precision Humanoid Joint Module Business Head Xu Peilin introduced that previously domestic reducers would grind teeth after running tests for a period, with precision degrading over use. Currently, Jiya Precision's RV reducers have reached internationally leading levels in precision, precision retention, reliability, and lifespan. Supporting this are three R&D centers in Foshan, Shanghai, and Osaka, Japan, plus Midea's over 30 years of motor technology accumulation.
天太机器人更是直接定义核心零部件。2023年，天太机器人牵头制定机器人核心零部件领域的国家标准《精密电驱动谐波齿轮模组通用技术规范》，并于2025年由国家市场监督管理总局正式发布。华沿机器人则从运动基座切入，发布具身智能运动基座，想用一底座多形态替代专机专用的开发逻辑。大模型解决的是认知问题，但小脑能不能精准执行、柔顺交互、稳定闭环，才是具身智能从演示走向产线的真正瓶颈。在大湾区这条具身智能机器人产业链上，整机、核心零部件、运动基座企业同向发力，共同补齐短板，最终指向同一个目标，让具身智能机器人不只是展台上灵动的样机，而是工厂里算得过来收益的生产工具。
Tiantai Robotics goes further, directly defining core components. In 2023, Tiantai Robotics led drafting of the national standard "General Technical Specification for Precision Electric Drive Harmonic Gear Modules" in the robot core component field, officially released by the State Administration for Market Regulation in 2025. Huayan Robotics enters from the motion base, releasing an embodied intelligence motion base aiming to replace "dedicated machine" development logic with "one base, multiple forms." Large models solve the cognition problem, but whether the cerebellum can precisely execute, compliantly interact, and stably close loops is the real bottleneck for embodied intelligence moving from demonstration to production lines. Along this Greater Bay Area embodied intelligence robot supply chain, complete machine, core component, and motion base enterprises pull together to fill shortcomings, ultimately pointing to the same goal—making embodied intelligence robots not just agile prototypes on exhibition stages but production tools whose returns calculate out in factories.
