All research

Robotics · Flagship research · 2026-08-22

A motor is not an actuator: inside the joint that makes a robot move

A robot joint forces motors, gears, sensors, bearings, heat, control and safety to agree inside a small housing. The hard part is not making it move once. It is producing the same controlled force, at the same cost and lifetime, thousands of times.

, Founder and publisher, AI Bottlenecks

A robot can look uncannily alive while the economics underneath it remain mostly unproven. A hand closes around a cup. A knee catches a falling body. An arm places a part with millimeter precision. The visible trick is motion. The less visible achievement is a joint that can repeat that motion without drifting, overheating, wearing out or bankrupting the bill of materials.

That joint is the actuator. It is often called the robot's muscle, which is useful up to a point. A muscle does not explain the gearbox, encoder, bearings, torque sensing, brake, power electronics, thermal path, housing, wiring, firmware and calibration packed around it. An actuator is closer to a small controlled machine inside the larger machine.

The distinction matters because the actuator thesis is frequently reduced to a parts count. Robots need many motors, the argument goes, so motor suppliers must win. That skips the engineering question that decides where value lands: which layer turns electrical energy into controlled, repeatable force at an acceptable weight, lifetime and cost?

Start with the command

A controller asks an elbow to move to a position, at a speed, while applying a particular amount of force. The motor creates rotation. A transmission reshapes that rotation into the torque, speed or linear force the joint needs. Bearings constrain the motion. An encoder reports position. A force or torque sensor reports load. The drive and control loop compare the command with what actually happened, then correct the error.

This happens continuously. If the joint encounters a box that is heavier than expected, the controller has to notice. If a human steps into its path, the joint has to respond safely. If friction changes as the mechanism warms, the control loop has to compensate without becoming unstable.

The useful output is not rotation. It is controlled force.

The common stack inside a joint

Open a serious actuator and the neat categories blur. The motor housing may also be the robot's structural member. A bearing arrangement can affect encoder accuracy. Gear friction becomes a thermal problem, a battery problem and a control problem at the same time.

Rotary and linear architectures send money to different places

Most robotics baskets speak as if an actuator were one standard object. It is not. Two broad motion families matter.

In a rotary joint, a motor turns an output around an axis. A precision reducer can trade motor speed for joint torque. Shoulders, elbows, hips and wrists often use this family.

In a linear actuator, a motor turns a screw that converts rotation into a push or pull. The linkage then turns that linear movement into joint rotation. Knees, ankles and compact electric cylinders can use this approach.

The choice is not cosmetic. A strain-wave joint rewards expertise in thin, elastic gear elements and compact zero-backlash behavior. A cycloidal design rewards rigidity, shock tolerance and load capacity. A screw-driven joint rewards precision thread geometry, nut life, lubrication and linear bearings. A tendon or quasi-direct design may reduce gearbox content, but it raises the burden on motors, routing, control and calibration.

One architecture decision can move the profit pool from reducers to motors, screws, sensors, software or integration. This is why any claim about universal content per humanoid deserves suspicion until a production architecture is fixed.

Why use a reducer at all?

Electric motors are often happiest spinning faster than a robot joint should move. A reducer lowers output speed and multiplies usable torque. That lets a smaller motor do useful work at the joint.

The bargain introduces its own problems.

The reducer is not merely there to make the motor stronger. It has to multiply torque without destroying control.

Harmonic Drive Systems sells strain-wave gearing and compact rotary actuators across a wide range of sizes. Its product catalog makes the architectural breadth visible: component sets, gearheads and integrated actuators are different levels of the same stack. Nabtesco specializes in high-rigidity precision reducers used in the heavier joints of industrial robots. Kollmorgen's frameless motor line is designed to disappear into the OEM's mechanism rather than arrive as a complete external motor.

Those positions can all be valuable, but for different reasons. The investor's job is to identify which capability remains hard to substitute after the OEM has chosen an architecture and begun cost-down work.

The actuator is a bundle of conflicts

The joint has to be small enough to fit the body and light enough not to punish every upstream joint. It has to be strong enough to move payload, efficient enough to preserve runtime and precise enough to manipulate objects. It may need compliance or backdrivability for safe contact. It has to survive falls and repeated cycles. It also has to become cheap enough to repeat across the machine.

Improving one variable can hurt another.

A larger motor can make more torque, but adds mass. More reduction can make torque easier, but may increase reflected inertia, friction and wear. A stiffer housing can improve accuracy, but adds weight. Aggressive cooling can protect continuous performance, but consumes space and power. Tighter part tolerances can make calibration easier, but raise machining cost and scrap.

The bottleneck is the compromise, not any single component.

This creates three separate tests for a public-market thesis.

Technical scarcity

Can the module deliver the required torque, precision, efficiency, compliance and lifetime? A clever prototype is evidence, but repeatable performance across temperature and duty cycle is stronger evidence.

Supply scarcity

Can qualified parts and finished modules be produced quickly at stable yield? Component nameplate capacity says little if assembly variation, test time or calibration holds back accepted output.

Economic scarcity

Can the supplier retain value after OEM insourcing, Chinese cost-down and architecture changes? A part can be technically impressive without being scarce. It can be scarce without becoming a good business.

Who owns the robot muscle?

No single company does.

The robot OEM owns the architecture, system-level control and final qualification. Specialist suppliers own pieces of the motion stack. Harmonic Drive Systems has deep strain-wave and compact-actuator capability. Nabtesco is established in high-rigidity precision reducers for medium and large industrial robot joints. Kollmorgen sells frameless motors for tight mechanical integration. Schaeffler is trying to integrate across motors, reducers, bearings, sensors, drives and industrial manufacturing.

Schaeffler's February 2026 humanoid presentation is a useful receipt precisely because it is modest. The company reported 32 sample orders and one serial order. That is commercial evidence, but not yet proof of material revenue. In January 2026 it also announced a partnership with Humanoid that includes actuator supply and plans to deploy several hundred humanoids in Schaeffler factories over five years. The arrangement gives Schaeffler both a supplier role and a place to learn from field use.

Nabtesco's Q1 FY2026 results show that precision-reducer demand is already economically real outside the humanoid story. Precision-reducer sales rose by JPY 4.8 billion year over year. Its Q1 question-and-answer document said the strong order environment was expected to continue, while also flagging uncertainty around North American automotive investment. That is an important distinction. Industrial robot recovery can tighten the same production system before humanoids contribute much revenue.

The right conclusion is not that one of these firms owns every future robot joint. It is that existing industrial-motion incumbents have manufacturing receipts, customer qualification processes and field histories that newer entrants still have to build.

Architecture can kill the basket

A robotics basket can be directionally right about robot volumes and wrong about component economics.

If OEMs converge on integrated rotary actuators, value may pool around reducers, frameless motors, bearings and module integration. If screw-driven linear joints take more of the lower body, screw makers and linear-motion specialists gain content. If quasi-direct or tendon systems work at scale, precision gearbox content may fall while motors, power electronics and control become more important.

Insourcing matters too. A robot maker may buy specialist components but design the module, housing, sensing and controls itself. Another may buy a complete actuator. Those two procurement choices create different supplier revenue from the same robot count.

The correct unit of analysis is therefore not robots shipped. It is the qualified architecture, the make-or-buy boundary and the economic value of each retained layer.

The factory may be the hidden bottleneck

Once the parts arrive, every finished actuator still has to be characterized.

Does commanded torque match delivered torque? Does the encoder agree with the real joint angle? How much friction, noise, heat and backlash sit inside the module? Does it pass burn-in? Can the factory trace an outlier back to a motor, gear, sensor, assembly step or firmware configuration?

This is where a component story becomes a manufacturing-system story. A factory needs fixtures, metrology, end-of-line test, calibration software and a data record for every module. Test takt time can limit output even when machining capacity is available. Rework can consume scarce benches and skilled technicians. A poorly traced field failure can turn into a broad and expensive containment exercise.

The calibration bottleneck deserves its own teardown, because it may decide how quickly actuator capacity becomes qualified robot output.

What would prove the thesis?

The actuator thesis does not need another backflip. It needs a receipt ladder.

  1. Qualified design: a named platform selects the component or module for a defined architecture.
  2. Sample order: the customer pays for hardware and begins formal evaluation.
  3. Serial order: the supplier has a production award with delivery expectations.
  4. Recognized revenue: humanoid or mobile-robot demand becomes visible in reported sales rather than a slide deck.
  5. Stable module yield: production volume rises without a matching rise in scrap, rework or test time.
  6. Field reliability: uptime, warranty cost and service intervals hold under repeated work.
  7. Repeat orders: the same customer or application expands after real deployment.

The negative evidence is just as concrete. A major architecture shift can remove a component. Persistent calibration bottlenecks can cap accepted output. Falling price without falling manufacturing cost can erase the margin pool. OEM insourcing can turn a supplier from a system partner into a commodity part vendor. Field failures can expose a lifetime assumption that no bench test caught.

The read-through

The useful actuator thesis is not that robots need lots of motors. It is that useful robots need controlled force, and controlled force is a systems problem.

As robots leave demos and enter repeated work, value should migrate toward the companies that can deliver that control at low weight, stable yield, predictable life and falling cost. Some will own a hard component. Some will own the integrated module. Some will own the fixtures, measurement and production knowledge that let every module behave the same way.

The robot is the visible product. The repeatable joint may be the real manufacturing achievement.

Sources and receipts

This research is educational and is not investment advice.