system
Muscle covers the actual moving parts of a robot's joints: actuators, reducers, encoders, and hands. These parts make up a large share of a robot's total cost, so as robot production grows, that growth converts most directly into revenue for whoever makes them.
执行器、减速器、编码器与手。每关节含量是人形放量最直接的传导路径。
Bottlenecks瓶颈 P13 Dexterous manipulation灵巧操作 · P14 Actuators, reducers, encoders执行器、减速器与编码器
What it isMuscle is where a robot actually moves and touches things. On one side: dexterous manipulation and end effectors, the hands, grippers, and tools at the end of a robot's arm that do the actual gripping and handling (P13, Dexterous manipulation). On the other: the actuators, reducers, and encoders that give a robot's joints torque, precision, and durability (P14, Actuators, reducers, encoders).
是什么“肌肉”（Muscle）是物理执行层：一端是灵巧操作与末端执行器（P13，灵巧操作），另一端是赋予机器人关节扭矩、精度与耐久性的执行器、减速器与编码器（P14，执行器、减速器与编码器）。
ThesisMuscle's capital-efficiency rank is strong for P14, Actuators, reducers, encoders: it rose four places on the CE-N re-rank to 4th of 18, among the best on the board. The argument behind that rank: these are the parts that make up the most expensive share of a robot's joints, so as humanoid robot production grows, that growth should convert most directly into revenue for whoever makes them. But the market has already priced in a lot of that growth before it's actually happened: Leaderdrive, a reducer and actuator maker, trades at roughly 470x P/E and 100x EV/Sales on humanoid demand that hasn't materialized yet. And even if that demand does arrive, it only turns into revenue for a supplier that can actually win orders in a reducer supply chain already running at capacity on non-humanoid demand, which is exactly why humanoid robot makers like Figure, Unitree, and Apptronik are building their own actuators in-house instead of waiting for outside suppliers.
论点该分类法对 P14，执行器、减速器与编码器 的论点是：每关节含量是人形机器人放量向零部件收入传导最直接的路径，据此它在 CE-N 再排名中上升四位，至 18 个瓶颈中的第 4 位。但承载这一论点的上市代理公司，早已把大量尚未到来的人形放量计入了股价：绿的谐波市盈率约 470 倍、EV/Sales 约 100 倍，押注的是一场尚未发生的人形渗透。每关节含量唯有在供应商能于已被非人形需求占满的减速器供应链中拿到订单份额时，才能转化为收入——这也是 Figure、Unitree 和 Apptronik 纷纷自建执行器、而不愿坐等这份配额的原因。
Products mapped to this system映射到本系统的产品
Multi-fingered robotic hands and hand-like end effectors; the conference-doc taxonomy's largest sub-category, spanning humanoid-hand specialists and teleoperation-data hand builders.
1 companies 1 家公司 +15 named in conference materials, unverified +15 项来自会议材料，未经核实
Grippers and end-of-arm tooling for industrial and collaborative manipulation, including electric/rotary grippers, electric cylinders, and voice-coil actuators.
1 manufacturers 1 家制造商 +2 named in conference materials, unverified +2 项来自会议材料，未经核实
Integrated joint actuators and harmonic/planetary/linear drive modules named in the conference-doc taxonomy; none of the named companies in this sub-category were found elsewhere on the site under any system.
Reducers & Gearing减速器与齿轮传动 · Servo Motors & Drives伺服电机与驱动
+11 named in conference materials, unverified +11 项来自会议材料，未经核实
Motors, drives, and gear systems feeding robot joints, including precision harmonic, planetary and RV robot reducers and industrial/household-appliance geared motors.
Servo Motors & Drives伺服电机与驱动 · Reducers & Gearing减速器与齿轮传动
1 companies 1 家公司 +1 named in conference materials, unverified +1 项来自会议材料，未经核实
Companies mapped to this system映射到本系统的公司
Manufacture mapped to this system映射到本系统的制造
The clearest read on actual demand comes from the two dominant harmonic- and cycloidal-reducer incumbents, both of which report on a lag of months, not quarters. Harmonic Drive Systems raised full-year operating-profit guidance 37% in its August 2026 print, and Nabtesco’s Component Solutions orders rose 26% year over year with its Changzhou plant running at 100% utilisation. Neither company named humanoids as a driver — both attributed the upcycle to semiconductor manufacturing equipment, machine tools, industrial robots, and Chinese EV makers.
That cuts two ways: it means humanoid volume is not yet visible in the order books of the companies best positioned to see it first, and it means that whenever humanoid volume does arrive, it arrives into a reducer supply chain that is already full — which is why Figure, Unitree, and Apptronik are all building their own actuators in-house rather than waiting for merchant allocation.
Leaderdrive, the sector’s most expensive listed proxy at roughly 470x P/E on FY2025 net profit of RMB124m, is the clearest case of a stock pricing a volume ramp the incumbents’ own order books do not yet show.
The vertical-integration counter-case has already been run once, and it failed. ESTUN — a Chinese robot maker that also makes its own components, marketed as the test of whether component self-supply rescues body-level gross margin — posted a 0.93% net margin in 2025 on RMB4.9bn of revenue, declining revenue in Q1 2026, and an H1 2026 headline of “+2,145% to +2,594% profit growth” of which the recurring portion was only RMB60-75m. Component self-supply did not rescue the margin.
Sell-side coverage of the actuator bill-of-materials share is unusually well corroborated — at least five houses (J.P. Morgan, Barclays, UBS, Korea Investment & Securities) converge on actuators, reducers, and encoders at 30-60% of a humanoid’s BOM. Corroboration stops there: 2030 shipment forecasts from the same houses span 1.75 million to 76 million units, a nearly 40x range that makes any single volume number unusable as a sector consensus.
On Leaderdrive specifically, a Deutsche Bank note claims the company already books RMB100m of humanoid-specific revenue — a claim that directly conflicts with the absence of any such disclosure in Leaderdrive’s own filings, and needs primary-source resolution before it can be trusted either way. The clearest illustration of how little of this is fundamentals-driven is Hyundai Mobis: UBS downgraded it to Neutral in June 2026 after a sharp rally, then J.P. Morgan upgraded it to Overweight five weeks later, both calls built on essentially the same 2030E actuator-revenue estimate — the stock, not the number, moved.
A wider bench of Chinese component names (Sanhua, Hengli Hydraulic, Minth, Huayan Robotics, Rokae) now guide explicit humanoid-revenue figures, but some peers covered in the same reports have already stopped giving that guidance — itself a signal about how far ahead of realised revenue the guidance had run. Separately, UBTECH’s disclosed manufacturing-capacity claims escalate across sell-side notes without ever reconciling to each other, echoing the primary-evidence caution above that a capacity claim is not a shipment.
P14, Actuators, reducers, encoders’s capital-efficiency rank is real, but the transmission mechanism the taxonomy originally assumed — component order books as a leading indicator of humanoid demand — appears backwards. The order books are currently a proxy for the semicap and industrial capex cycle, correlated with the same cycle driving Teradyne’s Semiconductor Test growth, not a humanoid-specific signal.
And “content per joint” only converts to revenue for a supplier that gets allocation in an already-full supply chain — which the largest humanoid OEMs are visibly trying to avoid needing, by building the capability in-house. The Hyundai Mobis whipsaw is the same problem in different clothes: two banks reached opposite ratings from the same underlying number, which is a narrative-elasticity story, not a fundamentals story.
Avoid paying a hardware-cycle multiple for a humanoid-specific thesis; wait for a named humanoid customer inside an incumbent’s own order-intake disclosure, which as of this research cut does not exist at either Nabtesco or Harmonic Drive.
关于真实需求最清晰的读数，来自谐波与摆线两大减速器龙头，二者的财报披露都以月而非季度为滞后单位。哈默纳科（Harmonic Drive Systems）在 2026 年 8 月的财报中将全年营业利润指引上调 37%；纳博特斯克（Nabtesco）的精密传动业务板块订单同比增长 26%，其常州工厂产能利用率达到 100%。两家公司均未将人形机器人列为驱动因素——都将本轮上行归因于半导体制造设备、机床、工业机器人与中国电动车厂商。
这一点具有两面性：一方面说明人形机器人的放量尚未反映在最有条件率先看到它的公司的订单簿中；另一方面也说明，一旦人形放量真正到来，它将进入一条已经满负荷运转的减速器供应链——这正是 Figure、宇树科技与 Apptronik 都在自建执行器产能、而非坐等经销商分配的原因。
绿的谐波（Leaderdrive）是该板块估值最高的上市代理，按 2025 财年净利润 1.24 亿元人民币计算市盈率约 470 倍，是股价提前计入尚未在龙头企业订单簿中显现的放量最清晰的案例。
垂直整合的反例已经被验证过一次，且以失败告终。埃斯顿（ESTUN）——一家同时自产零部件的中国机器人厂商，被视为“零部件自供能否挽救本体级毛利率”这一命题的试验样本——2025 年净利率仅 0.93%，营收 49 亿元人民币，2026 年第一季度营收同比下滑，2026 年上半年“利润增长 2145% 至 2594%”的头条数字中，经常性部分仅为 6000 万至 7500 万元人民币。零部件自供并未挽救毛利率。
卖方对执行器 BOM 占比的覆盖出奇地一致——至少五家机构（摩根大通、巴克莱、瑞银、韩国投资证券）都认为执行器、减速器与编码器占人形机器人物料清单的 30% 至 60%。一致性仅止步于此：同样是这些机构对 2030 年出货量的预测，从 175 万台到 7600 万台不等，跨度接近 40 倍，使得任何单一出货量数字都无法作为行业共识使用。
就绿的谐波而言，德意志银行一份笔记声称该公司已经确认 1 亿元人民币的人形机器人专属收入——这一说法与公司自身财报中不存在此类披露直接冲突，需要一手信源核实后方可采信任一方。最能说明这一切有多不由基本面驱动的例子是现代摩比斯（Hyundai Mobis）：瑞银在 2026 年 6 月一轮急涨后将其下调至中性评级，五周后摩根大通又将其上调至增持评级，两次评级几乎建立在同一个 2030 年执行器营收预测之上——变的是股价，不是数字。
一批更广泛的中国零部件企业（三花智控、恒立液压、敏实集团、华研机器人、卓益得）如今给出明确的人形机器人营收指引，但同一批报告覆盖的一些同行已经停止给出该项指引——这本身就说明指引跑在已实现收入前面有多远。另外，优必选披露的产能声称在各家卖方报告之间不断攀升却从未相互印证，与上文一手证据中“产能声称不等于出货”的告诫相呼应。
P14，执行器、减速器与编码器 的资本效率排名是真实的，但本分类法最初假设的传导机制——零部件订单簿作为人形需求的领先指标——似乎方向反了。目前订单簿更像是半导体设备与工业资本开支周期的代理指标，与推动泰瑞达半导体测试业务增长的是同一个周期，而非人形机器人的专属信号。
而“每关节含量”只有在一条已经满负荷的供应链中获得分配的供应商身上才能转化为收入——而最大的几家人形机器人整机厂，正明显在试图通过自建产能来避免依赖这种分配。现代摩比斯的评级摇摆是同一个问题的另一种表现：两家银行基于同一个底层数字得出相反的评级，这是叙事弹性的故事，不是基本面的故事。
不要为一个人形机器人专属论点支付硬件周期式的估值倍数；应等待某家龙头企业自己的订单披露中出现具名的人形客户——截至本轮研究，纳博特斯克与哈默纳科都还没有这样的披露。