What Are Robot Actuators? Motors, Gears, Costs, and Companies

A robot actuator is the part of a machine that turns energy into controlled physical motion. It rotates an arm, bends a knee, closes a gripper, drives a wheel, extends a cylinder, or tensions a cable. Software may decide where the robot should go, but actuators create the force that makes the decision real.

The simplest mental model is muscle. The better engineering model is an entire joint organ. A modern electric rotary actuator can combine a motor, precision reducer, bearings, encoders, torque sensing, brake, power electronics, communications, thermal path, seals, lubrication, and structural housing. Calling that assembly a motor is like calling a car an engine.

This guide explains what sits inside an actuator, why different robots need different architectures, how many actuators common robot types use, which companies supply the stack, and what public prices can and cannot tell a buyer. It also explains why the actuator line in a robot budget is much larger than the catalog price of its motor.

An electric motor creates torque at its rotor. Many robots need more output torque and less speed than a compact motor naturally provides, so a transmission changes the relationship. Bearings carry the robot structure and payload. An encoder measures position. A servo drive regulates motor current. A controller closes motion loops. A brake can hold a vertical axis when power is removed. The housing aligns these parts while carrying heat into the robot frame.

Every layer changes the behavior of the others. A high reduction ratio multiplies motor torque, but it also reflects inertia, friction, and transmission error back into the joint. A compact housing saves space, but it may trap heat. A hollow shaft makes cable routing cleaner, but it consumes valuable diameter. A stronger output bearing can improve stiffness while adding mass that the next joint must move.

This is why actuator development is a systems problem. The best motor on a bench can become a poor robot joint after it is paired with the wrong reducer, poorly supported by bearings, measured by a noisy encoder, or placed inside a housing that cannot reject heat.

A higher level computer asks for a joint position, velocity, or torque. The local controller compares that request with encoder measurements. The servo drive adjusts current through the motor windings. Magnetic fields create rotor torque. The transmission converts that motion into the speed and torque available at the output. Bearings deliver the load into the robot structure, while sensors report what actually happened.

These loops operate at different rates. Perception and task planning can tolerate more delay than joint current control. A robot can therefore run a sophisticated model on a central computer while closing fast current and position loops near the actuator. Local control also reduces the amount of time critical traffic moving across the robot network.

Force control requires more than a command labeled torque. The machine needs a credible estimate of output force, known transmission behavior, enough control bandwidth, and a mechanical structure that does not hide impacts behind friction. Some systems use a dedicated torque sensor. Others estimate output torque from motor current and a calibrated transmission model. Both approaches have error sources.

Electric actuation dominates many contemporary robots because batteries, motors, drives, and digital control can be packaged into compact modules. It is not the only answer. Hydraulic systems remain attractive when force density and impact tolerance outweigh plumbing complexity. Pneumatic systems can be fast, inexpensive, and clean at the point of use when a facility already supplies compressed air. Cable, tendon, screw, and belt mechanisms often relocate or reshape the output of another actuator.

The visible motion and the energy source are separate classifications. A rotary electric motor can create linear motion through a ball screw. A hydraulic cylinder creates linear motion directly. A rotary pneumatic vane can turn a fixture. The buyer should describe the required output before selecting the energy source.

A reduction stage lets a relatively small, fast motor produce useful joint torque. The ratio is not free leverage. Higher ratios can increase reflected inertia and friction, reduce backdrivability, magnify transmission imperfections, and restrict output speed. The correct reducer is the one that fits the complete motion and load cycle.

Strain wave reducers are compact, precise, and commonly associated with robot wrists, compact arms, and collaborative systems. Cycloidal reducers are valued for rigidity, load capacity, and shock tolerance in larger industrial axes. Planetary gearsets can deliver efficient, compact reduction across many size and ratio classes. Screws convert rotation to linear motion, while belts and cables can relocate mass or isolate geometry.

Harmonic Drive Systems says a vertically articulated robot typically uses six precision reducers and that compact or collaborative robots commonly use five to six. Nabtesco estimates that its precision reducers hold about 60 percent of the market for joints in medium and large industrial robots. Both statements come from the suppliers and should be read as company disclosures, not independent market audits.

A degree of freedom is an independently controlled motion. A dedicated actuator often controls one degree of freedom, but the relationship is not always one to one. One actuator can drive several coupled joints. Several actuators can cooperate around one joint. Passive springs, differentials, linkages, and underactuated fingers can add motion without adding an equal number of motors.

For a conventional articulated arm, the count is straightforward. Universal Robots describes its current arms as six rotating joints, which means six servo joints in the arm before any gripper, external rail, turntable, or tool changer is added. A quadruped with three controlled joints per leg uses twelve leg actuators. A mobile base can use two wheel motors, four independent wheel drives, or more elaborate steering and suspension actuation.

Humanoids show why configuration matters. Unitree publishes 23 joint motors for the base G1 and a range of 23 to 43 degrees of freedom for G1 EDU configurations. Schaeffler says an average humanoid uses 25 to 30 actuators in major body joints. Hands can change the total dramatically. Harmonic Drive Systems notes that a humanlike hand with three reducer equipped joints per finger would use 15 reducers per hand, or 30 across two hands.

There is no single actuator price because the category stretches from a small educational smart servo to a custom industrial joint qualified for years of high duty operation. Torque class, volume, integration, sensing, precision, safety, environmental protection, documentation, and service support all change the number.

Public stores provide useful reference points. ROBOTIS listed its AX 12A smart actuator at $57.39, several X series units around $103 to $632, and larger DYNAMIXEL P models from about $1,092 to $3,542 when reviewed. CubeMars listed compact robotic actuation modules from about $135.90 and larger AK modules from roughly $299 to $990. These are retail snapshots before shipping, taxes, brackets, cables, power supplies, controllers, spares, and integration.

The public range does not reveal the cost inside a mass produced commercial robot. An original equipment manufacturer may buy components under confidential contracts, manufacture key elements internally, amortize tooling over volume, and use several actuator sizes across one body. Retail multiplication can illustrate sensitivity, but it cannot reconstruct a private bill of materials.

The actuator purchase price is only the beginning. Engineers must size the joint, model loads, design housings, route cables, tune controls, validate thermals, manage suppliers, build fixtures, inspect parts, and test life. Certification and safety engineering can be substantial for machines that share space with people or operate in regulated environments.

A cheap module can become expensive if it needs a custom adapter, separate output bearing, external drive, fragile connector, complex calibration, or frequent replacement. A more expensive integrated joint may lower development time and field service. The correct comparison is cost per dependable operating hour at the required performance, not dollars per newton meter on a product page.

Production volume changes the answer again. A research team values availability, documentation, flexible software, and the ability to buy one replacement. A robot maker building thousands of units values repeatability, supplier capacity, test automation, second sources, tooling control, and a cost curve that improves without quality drifting.

The actuator supply chain is layered. Some companies sell one critical element, such as a frameless motor or precision reducer. Others sell integrated modules. Large robot manufacturers may design the joint internally while purchasing magnets, bearings, semiconductors, sensors, and production equipment from outside suppliers. Product availability does not prove that a component is installed in a named robot.

The established industrial leaders and the newer legged robot ecosystem are converging. Harmonic Drive Systems and Nabtesco bring decades of precision reduction experience. maxon and Kollmorgen supply motors and motion systems. Schaeffler and Nidec are packaging motors, gearing, sensing, and control around humanoid requirements. ROBOTIS and CubeMars make integrated modules accessible to researchers and smaller builders. Moog brings hydraulic and electromechanical experience to demanding robotic motion.

Nabtesco is strongest where large robot joints need rigid, durable reduction. The company says its RV precision reducers hold an estimated 60 percent share in medium and large industrial robot joints. Its product catalog spans components, gearheads, actuators, positioners, and mobile drive units.

Harmonic Drive Systems is closely associated with compact strain wave reduction. Its 2025 report describes five to six reducers in compact and collaborative robots, six in vertically articulated robots, and many more potential units in dexterous hands and humanoid bodies. The strategic attraction is clear: greater robot dexterity expands content per machine.

Schaeffler is approaching the humanoid joint as an integrated motion platform. Its planetary unit combines a two stage gearbox, electric motor, encoder, and controller, with a published torque range of 60 to 250 newton meters. Nidec is presenting a broader component basket that includes planetary and strain wave reducers, frameless motors, thermal products, and sensor integrated gearing.

maxon combines motors, gearheads, sensors, controllers, and increasingly complete robotic joints. Its published Reachy 2 case identifies EC flat and ECX torque motors, GPX gearboxes, and encoders inside compact multiaxis joints. Kollmorgen focuses heavily on frameless motor kits that let robot makers build the motor directly into their own structure. That gives the original equipment manufacturer packaging control while leaving bearings, housing, sensing, and transmission integration to the joint designer.

A humanoid needs enough torque to support and accelerate its own body while keeping limb mass low. Weight added at the hand raises the burden on the wrist, elbow, shoulder, torso, hips, knees, and ankles. Every gram has a downstream cost in torque, battery use, structure, and impact energy.

The joint must also tolerate contact. A factory arm can operate behind guarding with a stiff transmission and carefully managed path. A humanoid is expected to walk, recover balance, touch objects, and sometimes work near people. Backdrivability, torque sensing, compliance, fast control, and predictable stopping become more important.

Heat is the quiet constraint. A humanoid can show an impressive peak motion for seconds, yet productive work demands repeated cycles. Closely packed joints heat one another. Covers restrict airflow. Battery voltage limits current. The meaningful question is not maximum torque in isolation. It is the motion the whole body can repeat for an hour without overheating or exhausting its energy reserve.

Hands multiply the problem. Tiny actuators must fit inside fingers or send force through tendons from the palm and forearm. More independent motion can improve manipulation, but it adds motors, reducers, sensors, wires, calibration, and failure points. A simple gripper may be commercially superior for a bounded task even when a five finger hand looks more versatile.

Peak torque is the most abused number in actuator marketing. It describes a short event under stated conditions, not the load a joint can sustain all day. Continuous torque depends on temperature limits and cooling. Rated speed may fall as torque rises. A meaningful operating point needs both torque and speed at the same time.

Output torque also needs context. Is the figure measured before or after the reducer? Is it rated, peak, stall, or theoretical? Does the published mass include the drive, output bearing, housing, cables, and brake? Is efficiency measured at the relevant speed and load? Does the actuator survive the shock load created by a foot strike or collision?

Accuracy, repeatability, resolution, and backlash are different. A fine encoder can report tiny increments even when the transmission has compliance or lost motion. A highly repeatable joint can return to the same location while still having absolute error. Buyers should ask for complete system measurements at the output, under load, after thermal stabilization.

Actuator failures often begin as interactions rather than single broken parts. A bearing loses preload, which changes gear alignment, which increases vibration and encoder error. A seal adds drag, which raises current, which increases heat, which ages insulation and grease. A cable intermittently opens at one pose, so the problem appears to be software until flex testing reveals the conductor failure.

Wear is not always visible as catastrophic breakage. Backlash can grow gradually. Torque sensing can drift. Lubricant can migrate. A brake can release more slowly. Connector resistance can rise. The robot may remain operational while control quality, energy use, noise, and accuracy deteriorate.

Good diagnostics compare several signals. Rising current for the same motion can indicate friction or payload change. Temperature rise can reveal cooling degradation. Encoder disagreement can expose looseness or calibration drift. Vibration signatures can identify bearing or gear damage. The value is not a dashboard full of data. It is a maintenance decision early enough to avoid an unsafe event or expensive secondary damage.

Consider Meridian Robotics, a fictional team designing a twelve actuator inspection quadruped. The robot must walk for two hours, carry sensors, climb stairs, tolerate occasional foot impacts, and remain serviceable by a small field team. This is an example, not a quotation or a description of any supplier's customer.

The team first considers a $300 integrated module. Twelve units equal $3,600 at list price. That number is attractive, but the module basket is not the leg system. The robot also needs mechanical links, feet, cables, power distribution, compute, battery, structure, sensors, protective covers, assembly fixtures, spares, software, testing, and a support plan.

Suppose the team budgets two spare actuators, $800 of joint brackets and cables, $1,500 for battery and power hardware, $2,500 for compute and sensing, $1,600 for structure and covers, and $2,000 for prototype machining and assembly. The direct prototype basket reaches $12,600 before engineering labor, failed iterations, shipping, taxes, tools, and life testing.

A $470 module would raise the twelve installed units plus two spares from $4,200 to $6,580, a $2,380 difference. That premium could be excellent value if it avoids redesign and improves field life. It could also be wasted if the higher priced unit is heavier, thermally mismatched, difficult to control, or no more durable in the real load cycle. The pilot must measure the intended work.

Buying an integrated module can shorten the path to a working prototype. The supplier has already combined several difficult layers and may provide firmware, communication tools, drawings, and replacement inventory. The compromise is less control over package geometry, thermal paths, sensing, cost structure, and future supply.

Building the joint provides architectural control and a path to differentiated performance. It also transfers integration risk to the robot maker. Frameless motors, bare reducers, bearings, encoders, drives, housings, lubrication, assembly tolerances, and end testing must all work as one product. The first custom joint is almost never cheaper than the first purchased module.

A common path is staged. Research begins with catalog modules. The team measures the real torque, speed, impact, temperature, and duty cycle. A custom joint appears only where the evidence shows that size, mass, performance, supply, or cost prevents the product from scaling. Other axes can remain on standard hardware.

Begin with the motion, not a preferred actuator. Define the payload, geometry, acceleration, speed, duty cycle, contact events, ambient conditions, and life target. Convert those requirements into joint torque and speed over time. Include the mass of the actuator and every structure it forces the upstream joints to carry.

Then test the operating map. A short no load demonstration proves very little. Run representative cycles until temperatures stabilize. Measure electrical energy, output motion, current, temperature, noise, and control error. Add realistic cables, covers, mounting stiffness, payload, and ambient temperature. Test impact and recovery within a safe fixture.

Finally, inspect the commercial system. Ask who owns firmware, how field updates work, which parts have lifecycle commitments, what test data accompanies each unit, and how a failed actuator is diagnosed. The actuator becomes a fleet problem the moment more than one robot leaves the lab.

Better actuators can make a robot lighter, stronger, more efficient, more precise, safer in contact, and easier to service. Integrated modules can compress development schedules. Standardized joint families can reduce spare inventory and simplify control. Rich sensing can turn the actuator into a source of machine health data.

The same integration creates concentration risk. One compact module may contain several single source elements. A controller failure can require replacement of the entire joint. Proprietary firmware can limit tuning or repair. A beautiful specification can hide a narrow thermal envelope. Greater sensor count can create calibration and data quality burdens.

Actuators also cannot rescue a poor robot architecture. Excess reach, heavy distal links, weak cable routing, unrealistic payload targets, and badly chosen duty cycles will consume any gain from a better motor or gearbox. Mechanical design, controls, energy, and application scope must improve together.

Actuators are where a robot's intelligence meets physics. They are not commodity motors attached to clever software. They are compact motion systems whose electromagnetic, mechanical, electronic, thermal, and manufacturing choices determine whether the robot can do useful work repeatedly.

For buyers, the most important distinction is between peak capability and productive capability. A robot needs enough joint performance after heat, wear, payload, impact, and service constraints are included. The winning actuator is not the one with the largest torque number. It is the one that completes the required task with acceptable energy, accuracy, life, safety, availability, and total cost.

For investors and operators, actuator economics reveal why robot cost does not fall as quickly as compute cost. More dexterity means more joints. More joints mean more precision components, assembly steps, calibration, potential failures, and replacement inventory. Scale can lower prices, but quality must scale with it. The companies that combine repeatable manufacturing with validated motion performance will capture a disproportionate share of the physical AI value chain.

For robot builders, the sensible path is measurement before customization. Use available modules to learn the real operating envelope. Build custom hardware only where the application proves that catalog products block performance, packaging, supply, or unit economics. Motion is too central to optimize from a spreadsheet alone.