大数跨境

Foresight|减法先行:人形机器人商业化的常识回归

Foresight|减法先行:人形机器人商业化的常识回归 中国电子信息产业发展研究院
2026-11-16
0


《Foresight》英文专刊中,《减法先行:人形机器人商业化的常识回归》一文叩问了一个被资本热情暂时遮蔽的问题——当“全能劳动力”叙事席卷行业,率先跨过商业死亡谷的,为何恰恰是主动“做减法”的专用人形机器人?


In the English-language special issue of Foresight, the article “Subtraction First: A Return to Common Sense in Humanoid Robot Commercialization” raises a question temporarily obscured by capital momentum, as the narrative of the “universal workforce” gains traction across the industry, why are the first humanoid robots to traverse the commercialization chasm precisely specialized variants that proactively “pursue subtraction”?



2025年,全球近300家人形机器人企业发布约400款新品,头部企业屡获大额融资。但同年全球人形机器人总出货量仅约2万台,发布会上的惊艳的“全能”演示,难以直接转化为可付费的刚需订单——市场因通用路线而火热,客户却只为专用路线的确定性付费。


In 2025, nearly 300 humanoid robotics companies worldwide launched approximately 400 new products, with leading firms securing multiple large-scale funding rounds. Yet the total global shipment volume of humanoid robots reached only around 20,000 units in the same year. The impressive “universal-capability” demonstrations showcased at product launches cannot be readily translated into paid, mission-critical orders. While the market gains momentum around the general-purpose development path, customers make purchasing decisions based on the reliability offered by the specialized path.







通用的隐性成本:“全能”成为商业化的减速带

Hidden Costs of the General-Purpose 

Development Path: When All-Round 

Capabilities Become a Commercial Decelerator


当前,通用路线隐性成本内生于技术路线本身,且在商业化进程中被持续放大:


At present, the implicit costs of the general-purpose pathway are inherent to the technical approach itself and continue to be amplified throughout the commercialization process.


数据门槛横亘在规模化之前。通用机型的数据需求与采集现实之间存在矛盾。头部企业实验显示,10万次真实数据训练后的抓取任务成功率仅58%,10亿次后才跃升至87%,而通用路线需万亿级数据量。


A data threshold stands ahead of scale. A gap exists between the data requirements of general-purpose models and the realities of data collection. Experiments conducted by leading firms indicate that the success rate for grasping tasks reaches only 58% after training on 100,000 real-world data samples, rising to 87% only after one billion samples. The general-purpose pathway requires data volumes at the trillion-sample scale.


开放性任务让可靠性失守。人形机器人越无所不能,自由度越多,失效模式就越多;感知模态越复杂,误触发概率就越高;大模型泛化能力越强,行为边界就越不可预知。海量未知任务下,整机可靠性的下限由最脆弱的环节决定。特斯拉因其低可靠性放弃当年量产数千台Optimus的计划。


Open-ended tasks pose challenges to reliability. The greater a robot's degrees of freedom and the more complex its perceptual modalities, the higher its probability of false triggering and multi-node failure. Furthermore, as the generalization capability of the foundational model increases, its behavioral boundaries become inherently less predictable. When confronted with a large set of unknown tasks, the minimum level of overall system reliability is determined by its weakest component. Tesla set aside its plan to mass-produce several thousand Optimus units that year due to reliability considerations.


解释性稀缺让责任无从安放。目前,通用路线通常采用端到端模型,当机器人突然改变抓取策略,整机厂与运营方无法回答“它为什么这么做”。企业无法划定风险边界、测算停机损失,批量采购的意愿便随之瓦解。


Scarcity of interpretability creates challenges for accountability. Current general-purpose approaches typically employ end-to-end models. When a robot unexpectedly shifts its grasping strategy, system integrators and operators cannot answer the question “why it took this action”. Without the ability to define risk boundaries and quantify downtime losses, enterprises see reduced willingness for bulk procurement.


成本刚性换不来产出。通用机型BOM约5万—10万美元,而任务收敛的专用机型可压至2万—4万美元。对只需搬运标准货箱的仓库而言,为叠衣服和爬楼梯支付的溢价,换不来任何可量化的产出增益。


Cost rigidity does not translate into productivity gains. General-purpose models have a BOM cost of approximately USD 50,000–100,000, while task-converged specialized models can bring this figure down to USD 20,000–40,000. For warehouses that only require standard-container handling, the premium incurred for garment folding and stair-climbing capabilities yields no quantifiable productivity gains.







专用的飞轮效应:“减法”如何加速商业闭环

The Specialization Flywheel: How a 

Subtraction Strategy Accelerates 

the Commercial Feedback Loop


专用路线的闭环,可以用采购最原始的标尺来衡量,当分母足够大、分子足够小时,商业循环自动成立:


A dedicated route's closed loop can be assessed against the most fundamental procurement yardstick: when the denominator is sufficiently large and the numerator sufficiently small, the business loop closes on its own.




投资回报周期=采购成本÷(年度节省人力成本+年度良率提升收益−年度维护费用)


Payback Period = Total Procurement Cost (CapEx) ÷ [Annual Labor Cost Savings + Annual Revenue from Yield Improvement − Annual Maintenance Expenses (OpEx)]




压小分子。以关节模组为例,专用机型去除多余性能冗余,总成本可由数十万美元降低至数千美元,通过规模化摊薄,成本曲线呈现指数级下降趋势。


Reduce the numerator. In the case of joint modules, dedicated machines eliminate redundant performance headroom, bringing total cost down from hundreds of thousands of dollars to several thousand. With costs amortized across scale, the cost curve declines exponentially.


做大分母。专用路线任务边界收敛使得人形机器人系统故障可枚举、决策链路具备较强可追溯性。同时,随着专用路线可靠性及稳定性的提升,专用机型的收益提升,维护费用降低,ROI可量化。


Enlarge the denominator. As task boundaries converge on the dedicated-machine route, faults in humanoid robot systems become enumerable, and decision chains exhibit strong traceability. At the same time, as the reliability and stability of the dedicated route improve, revenue increases, maintenance costs decrease, and ROI becomes quantifiable.







订单已经投票:专用的确定性正在兑现

Orders have cast their vote: the predictability 

of specialized solutions is being realized.


截至2025年底,真实进入场景的订单中,工业制造占据绝对主导,以汽车总装与3C精密装配为核心,其次是物流仓储与导览导购。


By the end of 2025, among orders deployed in real-world environments, industrial manufacturing accounts for an overwhelming share, centered on automotive final assembly and 3C precision assembly, followed by logistics and warehousing, and visitor guidance and retail assistance.


工业制造是专用路线最先取得商业突破的领域。人形机器人在工业领域已从概念验证阶段迈入小规模商业部署,瑞典Hexagon等多家整机厂已将人形机器人送入真实产线,并收获规模化订单。(关于专用人形机器人在工业制造场景的“减法”实践,请扫描前文图中二维码阅读原文)


Among all addressable application scenarios, industrial manufacturing is the first to achieve commercial breakthroughs via the subtraction strategy. Humanoid robots in industrial settings have advanced from proof-of-concept into small-scale commercial deployment. Multiple system integrators including Sweden’s Hexagon have deployed humanoid robots onto live production lines and secured volume orders. (For further reading on the “subtraction” practice of specialized humanoid robots within industrial manufacturing scenarios, please scan the provided QR code to read the full report.)


物流仓储正成为最具潜力的放量入口。环境结构化使单仓验证成果可快速复制至物流全网节点,RaaS租赁模式则更进一步将客户的试错门槛从一次性资产投入降为按月付费。


Logistics and warehousing is emerging as the most promising entry point for volume scaling. Structured environments enable outcomes validated within a single warehouse to be rapidly replicated across network-wide logistics nodes. The RaaS leasing model further lowers clients’ trial threshold by shifting capital expenditure from one-off asset investment to monthly subscription payments.


导览导购是通往服务场景的最佳跳板。导览导购人机交互服务边界相对可收敛、容错度高,在此积累的公众信任与商业经验,是人形机器人走向家庭、医疗等开放场景的入场券。


Guidance and retail assistance serve as the optimal stepping stone toward service use cases. This use case features relatively well-bounded human-machine interaction and high fault tolerance. Public trust and commercial experience accumulated here form the entry ticket for humanoid robots to expand into open environments such as households and healthcare.






通用智能是人形机器人的终局愿景,专用可靠是其必经之路。减去用不上的功能,换来的不是能力的退让,而是更低的成本、更高的可靠性与算得清的回报——减法先行,从来不是退步,而是一种以慢为快的战略清醒。


General intelligence stands as the end-state vision for humanoid robots, while specialized reliability constitutes its necessary progression. Removing unused capabilities does not represent a retreat in functionality; instead, it delivers lower costs, higher reliability and quantifiable returns. Prioritizing subtraction is never a step backward. It reflects strategic clarity: moving faster by taking a measured pace. 


《Foresight》是赛迪顾问首份全英文专刊,致力于精准开展前沿技术预测预见,密切追踪全球产业变革中的关键命题,以专业、独立、前瞻的研究,为全球产业界、投资界与政策制定者提供来自中国的洞察与判断。


Foresight is CCID Consulting’s inaugural English-language publication. Dedicated to delivering actionable intelligence on frontier technologies and deep structural tracking of global industrial shifts, Foresight delivers independent, authoritative, and forward-looking research—providing multinational executives, institutional investors, and global policymakers with strategic clarity and actionable perspectives from China.


推荐阅读


洞见商业化前沿——赛迪顾问全英文专刊《Foresight》正式发布

Foresight|规模的隐性成本:盲目扩容大模型为何损害商业落地可行性

Foresight|量子优越性之后,寻找第一个可持续的行业付费场景


来源 | 赛迪顾问

编辑 | 办公室


【声明】内容源于网络
0
0
中国电子信息产业发展研究院
中国电子信息产业发展研究院(英文简称CCID,赛迪)是工业和信息化部直属单位,是我国工业经济、信息技术和互联网等领域的著名智库,也是具有广泛影响力的公共技术服务机构和媒体传播集团。欢迎通过本平台与我们合作交流!
内容 2691
粉丝 0
中国电子信息产业发展研究院 中国电子信息产业发展研究院(英文简称CCID,赛迪)是工业和信息化部直属单位,是我国工业经济、信息技术和互联网等领域的著名智库,也是具有广泛影响力的公共技术服务机构和媒体传播集团。欢迎通过本平台与我们合作交流!
总阅读43.8k
粉丝0
内容2.7k