先说结论：机器人量产，先卡的往往不是大脑，而是身体Main Point: Humanoid Volume May Be Blocked by the Body Before the Brain
模型和控制很重要，但批量交付还要靠关节、减速器、传感器、线束、散热和制造良率。Models and control matter, but volume delivery also depends on joints, reducers, sensors, harnesses, thermal design, and manufacturing yield.
人形机器人最吸引眼球的是视频：走路、搬箱子、握手、进工厂。但投资研究不能停在视频。真正量产时，一台机器人要反复依赖几十个关节、执行器、减速器、传感器和线束。Humanoid robots attract attention through videos: walking, moving boxes, shaking hands, entering factories. But investment research cannot stop at video. Volume production depends on dozens of joints, actuators, reducers, sensors, and harnesses.
如果其中一个关键零部件良率不稳、成本太高、供给太集中，整机厂的量产节奏就会被拖慢。机器人故事的窄门，可能藏在身体里。If a key component has unstable yield, high cost, or concentrated supply, the OEM's production cadence slows. The narrow gate in humanoids may be inside the body.
先认人：一台机器人至少有五条供应链Cast: A Humanoid Has at Least Five Supply Chains
| 供应链 | 关键部件 | 要验证什么 |
|---|---|---|
| 关节 | 电机、减速器、编码器 | 寿命、噪音、成本、良率 |
| 传感 | 视觉、力矩、触觉 | 是否进入主流方案 |
| 算力 | 边缘芯片、控制器 | 功耗和实时性 |
| 能源 | 电池、线束、热管理 | 续航和安全 |
| 制造 | 校准、装配、测试 | 能否从试产走向量产 |
第一条是执行器和关节，包括电机、减速器、编码器和力矩传感器。第二条是感知，包括摄像头、深度传感器、IMU、触觉和力反馈。第三条是算力和控制。第四条是电池、线束和热管理。第五条是整机制造与校准。The first chain is actuators and joints: motors, reducers, encoders, and torque sensors. The second is sensing: cameras, depth sensors, IMUs, touch, and force feedback. The third is compute and control. The fourth is batteries, harnesses, and thermal design. The fifth is assembly and calibration.
如果只看整机品牌，很容易错过真正的卡点。Serenity 式读法会问：哪一个零件如果断供，整机厂不能简单换掉？哪一个零件要长期认证、良率爬坡和成本下降？If you only watch the OEM brand, you can miss the real bottleneck. The Serenity-style question is: which part cannot be easily replaced if supply breaks, and which part requires long qualification, yield ramp, and cost reduction?
故事开场：视频证明会动，供应链证明能卖Opening: Videos Prove Movement; Supply Chains Prove Saleability
一段视频能证明机器人会动，但不能证明它能被稳定生产。量产需要每个零件都有供应商、成本曲线、测试标准和售后可靠性。一个手指、一个髋关节、一颗力矩传感器，都可能把演示机和量产机隔开。A video proves the robot can move; it does not prove it can be produced reliably. Volume needs suppliers, cost curves, test standards, and service reliability for every part. A finger, hip joint, or torque sensor can separate a demo unit from a production unit.
这就是为什么机器人文章要从“谁做整机”继续追到“谁做身体”。如果量产真的来临，最先被放大的不一定是整机毛利，而是那些被多家整机厂同时需要的通用部件。That is why humanoid articles must move from who builds the robot to who builds the body. If volume arrives, the first amplified layer may not be OEM margin, but common parts needed by multiple OEMs.
证据路径：从试点客户，追到零部件认证Evidence Path: From Pilot Customers to Component Qualification
机器人供应链的硬证据，是试产数量、关键零件定型、良率和客户复购，而不是单次演示。Hard humanoid evidence is pilot volume, component design-in, yield, and repeat orders, not a single demo.
人形机器人链条还早，所以证据更要谨慎。McKinsey 和行业资料都把减速器、执行器、传感器和制造校准列为关键环节。文章要继续问：哪些部件已经被整机厂定型，哪些只是样机供应，哪些能跟随多家 OEM 走向量产。The humanoid chain is still early, so evidence needs care. McKinsey and industry sources point to reducers, actuators, sensors, and manufacturing calibration as key layers. The article should ask which parts have been designed in, which are only prototype supply, and which can follow multiple OEMs toward volume.
如果某个零件只出现在单一演示里，证据弱。如果它被多家整机厂采用，且量产计划和扩产动作能互相印证，证据才会变硬。If a component appears only in one demo, evidence is weak. If it is adopted by multiple OEMs and volume plans match capacity expansion, evidence strengthens.
反方：机器人最容易把试点当量产Bear Case: Robotics Often Mistakes Pilots for Volume
机器人赛道最大的反方，是时间。试点可以很快，量产很慢；视频可以好看，售后很难；客户可以试用，不一定会复购。供应链公司也可能先扩产，结果订单没有按预期来。The biggest bear case is time. Pilots can move fast, but volume is slow; videos can look good, while service is hard; customers can trial without reordering. Suppliers may expand before orders arrive.
另一个风险是设计变化。人形机器人还在快速迭代，今天的关节方案、传感器组合、手部设计，明年可能被改掉。被设计进去，不等于永远锁住。Design change is another risk. Humanoids are still iterating quickly. Today's joint architecture, sensor suite, or hand design can change next year. Being designed in does not mean being locked forever.
学习 takeaway：看机器人，先画 BOM，再看整机Learning Takeaway: Draw the BOM Before Looking at the OEM
下次看到机器人推文，先别问哪家整机最帅。画一张 BOM：关节、减速器、传感器、算力、电池、线束、制造。然后问哪一层最难替代，哪一层能供应多家客户，哪一层有量产证据。Next time you see a humanoid post, do not first ask which OEM looks coolest. Draw a BOM: joints, reducers, sensors, compute, batteries, harnesses, manufacturing. Then ask which layer is hardest to replace, which can supply multiple customers, and which has volume evidence.
这样你看到的就不是科幻视频，而是一条可以复核的工业链。Then you are not watching a sci-fi video; you are reading a verifiable industrial chain.