# Actuator

> Source: https://aiwiki.ai/wiki/actuator
> Updated: 2026-07-29
> Fact-checked: 2026-07-29
> Categories: AI Hardware, Robotics
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Actuator." aiwiki.ai, 29 Jul 2026. https://aiwiki.ai/wiki/actuator
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

An **actuator** is a device or mechanism that converts supplied energy and a command into mechanical force, torque, or motion. Actuators are the output elements of machines: a valve actuator turns or translates a valve stem, a cylinder pushes a load, and a robotic joint actuator produces motion about or along a joint. The word names a function, not a single technology. An actuator may be electrical, hydraulic, pneumatic, thermal, electrostatic, or based on another physical effect. It may produce rotary or linear motion, and it may be a simple cylinder or an integrated assembly containing a motor, transmission, sensors, brakes, and electronics. In [robotics](https://aiwiki.ai/wiki/robotics), the actuator is distinct from the controller that computes a command and from the sensor that measures the result, although all three may share one housing. ISO 8373:2021 is the published international vocabulary for robotics as of July 28, 2026; a revision is in development.[1]

Actuator selection is a system design problem. Required output force or torque, speed, stroke, duty cycle, thermal path, reflected inertia, compliance, efficiency, mass, available energy, environment, maintenance, failure behavior, and control bandwidth all matter. A claim that one actuation technology is categorically stronger, safer, more efficient, or more precise than another is therefore incomplete unless the load, motion profile, power supply, transmission, controls, test conditions, and system boundary are specified. A controlled experiment comparing one electric, one hydraulic, and one pneumatic linear system found the electric setup used the least power and followed the commanded motion most consistently, but the authors limited all three to 1.1 kW, omitted a hydraulic accumulator and pneumatic receiver, and tested particular components under a particular open-loop protocol. The result describes those rigs, not a universal ranking of actuator classes.[2]

## Functional boundary

An actuator lies between an energy source and a mechanical load. Its input can be electric current, pressurized fluid, heat, or another controllable energy flow. Its output can be force through a displacement, torque through an angle, or a constrained deformation. A complete actuation system can also include energy conversion upstream, such as a compressor or hydraulic pump, and power electronics or valves that meter the delivered energy. Whether those supporting components are counted as part of "the actuator" depends on the comparison being made. Product data may describe only a motor, a cylinder, a geared joint module, or an entire drive train. Meaningful comparisons keep that boundary consistent.[3]

A motor is not always a complete actuator. A rotary electric motor converts electrical energy into shaft torque and speed, but a machine may also need a gearbox, screw, cable, brake, bearings, position sensing, and structural interfaces before that output can move the load. Conversely, a direct-drive joint may use the motor as the principal mechanical output element. A hydraulic or pneumatic cylinder is itself an energy converter, but its usable motion still depends on valves, pressure supply, plumbing, sensing, and control. Soft and material actuators may combine the transducer and compliant structure so closely that a separate transmission cannot be identified.[3][4]

The controller and actuator also have different roles. In a conventional feedback loop, the controller compares desired and measured state and issues a low-power command. An amplifier, motor drive, or servo valve converts that command into a power flow. The actuator and transmission then act on the plant, while encoders, force sensors, pressure sensors, or other devices close the loop. This separation matters when a behavior is attributed to "the actuator." Tracking accuracy, contact stability, and disturbance rejection are properties of the coupled mechanism, sensors, controller, sampling, load, and environment, not of the energy converter alone.[4]

An actuator can be passive in one direction or hold a load without continuous commanded motion. A spring-return cylinder uses fluid power for one direction and stored elastic energy for the other. A brake can hold a joint after power is removed. Some screw mechanisms or transmissions resist backdriving because of friction and geometry, while others are readily backdrivable. These behaviors should be verified for the particular design. They are not guaranteed by the labels "electric," "linear," or "geared."[3][5]

## Mechanical output

The most direct actuator ratings describe output at a defined point. Linear output is specified by force, velocity, stroke, and often acceleration. Rotary output is specified by torque, angular velocity, and angular range or continuous rotation. For ideal translational and rotational motion, instantaneous mechanical power is:

$$
P = Fv
$$

$$
P = \tau\omega
$$

where \(F\) is force, \(v\) is linear velocity, \(\tau\) is torque, and \(\omega\) is angular velocity. These equations do not include losses or stored energy. Peak power can therefore differ substantially from average input power, especially when a spring, accumulator, pressure vessel, flywheel, or moving mass stores and releases energy.[2][3]

Work and energy ratings are also important for intermittent motion. Translational work is the integral of force over displacement, and rotational work is the integral of torque over angle. An actuator that can briefly produce high force may not be able to repeat that action at a high duty cycle because winding temperature, fluid temperature, seal friction, power-electronic limits, or structural stress accumulates. Manufacturers consequently distinguish continuous operating regions from intermittent or peak limits. In an electric motor, the torque constant relates current to developed torque under the assumptions used in the motor model, while resistance, speed, cooling, and the permitted winding temperature constrain continuous operation.[6]

For a fluid cylinder, pressure acting over effective piston area gives the idealized force contribution:

$$
F = pA
$$

Real net force must account for pressure on both sides of the piston, unequal areas in a single-rod double-acting cylinder, seal friction, and other losses. The rod reduces effective area on the rod side, so equal pressure does not normally produce equal extension and retraction force. Flow rate and effective area jointly determine an idealized piston velocity. Parker's cylinder engineering guide explicitly warns that simply multiplying a pressure difference by full piston area is incorrect for a single-rod cylinder when both chamber pressures matter.[5]

Output at the load is not necessarily output at the motor or cylinder. Lever arms, linkages, gears, screws, cables, and joint geometry transform force and motion. In a robot, the relationship between joint torque and force at an end effector also varies with configuration. A fixed motor torque can correspond to different endpoint forces at different poses, and a given endpoint motion can require different joint speeds. Quoting motor stall torque as if it were continuous joint torque, or quoting joint torque as if it were constant endpoint force, mixes distinct quantities.[6][7]

## Classification

Actuators can be classified along several independent axes. No single taxonomy captures every design.

| Axis | Common classes | What the classification describes |
| --- | --- | --- |
| Energy domain | Electrical, hydraulic, pneumatic, thermal, chemical, electrostatic | The supplied energy and primary conversion effect |
| Mechanical output | Rotary, linear, limited-angle, continuum deformation | The form of motion at the defined output |
| Transmission | Direct drive, geared, screw-driven, cable or tendon, linkage | How source motion and load motion are related |
| Compliance | Nominally rigid, passively elastic, variable stiffness, soft-bodied | How force changes with displacement and whether compliance is intentional |
| Control interface | Open-loop command, position servo, velocity servo, torque or force servo | The commanded variable at the chosen system boundary |
| Integration | Bare transducer, cylinder or motor, packaged drive, joint module | Which supporting components are included |

The axes should not be collapsed into one another. An electric actuator can be rotary or linear, direct-drive or highly geared, stiff or series-elastic, and open-loop or force-controlled. A hydraulic actuator can be a linear cylinder or rotary motor. A pneumatic device can be a rigid cylinder or a compliant artificial muscle. A soft actuator can be fluidic, electrically activated, thermally activated, or tendon-driven. The output form and control behavior do not follow uniquely from the energy source.[3][8]

Rotary actuators commonly drive revolute joints, wheels, drums, and valve shafts. Linear actuators drive prismatic joints, presses, positioning stages, and clamps. A [ball screw](https://aiwiki.ai/wiki/ball_screw), leadscrew, rack and pinion, belt, or cable can convert a rotary motor's output into translation. A rotary joint may likewise be driven remotely by a linear cylinder or tendon through a moment arm. The transmission changes speed, force, inertia, stiffness, backlash, efficiency, and packaging, so "rotary" and "linear" alone say little about overall performance.[3][9]

Actuator terminology can also reflect the controlled quantity. A position servo accepts a position target, but that does not mean its internal transducer is a distinct type of actuator. The same motor and transmission may operate under position, velocity, torque, or impedance control if sensing and electronics support those modes. Torque control based only on motor current is an estimate influenced by the torque constant, friction, acceleration of internal inertia, and transmission losses. A downstream torque or [force-torque sensor](https://aiwiki.ai/wiki/force_torque_sensor) measures at a different physical location and therefore observes different dynamics.[4][6]

## Electric actuation

Electric actuators commonly use brushed direct-current motors, [brushless DC motors](https://aiwiki.ai/wiki/brushless_dc_motor), alternating-current machines, stepper motors, voice coils, solenoids, or electrostatic devices. In robot joints, a brushless permanent-magnet motor with electronic commutation is common, but it is not universal. The drive regulates phase current or voltage, and an encoder or resolver may provide rotor and joint position. Motor current is useful for torque control because developed torque is approximately proportional to current within the applicable model, but magnetic saturation, temperature-dependent resistance, commutation, friction, and drive limits affect the realized behavior.[4][6]

Electric machines generally produce rotary motion, so transmissions are frequently used to trade speed for torque. For an ideal reduction with ratio \(G\), motor speed is divided by \(G\) and torque is multiplied by \(G\), while power is conserved. Real transmissions dissipate power through tooth contact, bearing friction, lubricant churning, seal drag, and deformation. The load inertia reflected to the motor decreases with the square of the ratio, while motor-side inertia reflected to the output increases with the square of the ratio. High reduction can therefore make a compact motor useful at a slow, high-torque joint, but it can also increase apparent inertia and friction at the load.[3][10]

Common rotary reductions include [planetary gear trains](https://aiwiki.ai/wiki/planetary_gear_train), cycloidal drives, and strain-wave gears. A strain-wave gear uses a wave generator, flexible externally toothed spline, and rigid internally toothed circular spline. The manufacturer Harmonic Drive describes the engagement and tooth-count difference that produce a large ratio in a compact package. Its catalog's zero-backlash statements apply to specified strain-wave products and operating conditions; they should not be generalized to every gearbox or to an entire joint, whose bearings, shafts, housings, and wear can add lost motion and compliance.[10]

Linear electromechanical actuators often pair a motor with a screw. A ball screw circulates rolling elements between screw and nut, reducing sliding friction compared with a plain leadscrew. Preload can remove axial clearance and increase rigidity, but excessive preload raises torque, heat generation, and life-related loads. Screw lead sets the ideal motion conversion, while critical speed, buckling, mounting, lubrication, contamination, and bearing capacity constrain an actual design. A [planetary roller screw](https://aiwiki.ai/wiki/planetary_roller_screw) uses threaded rollers rather than recirculating balls; its load capacity and life still depend on the particular geometry and duty cycle.[9]

Direct drive removes a discrete reduction between motor and load. It can reduce transmission friction, backlash, and reflected rotor inertia at the output, improving mechanical transparency. The cost is that the motor must produce load torque directly, which can increase motor diameter, mass, current, and thermal burden. A University of Pennsylvania study of small direct-drive legged robots emphasized robustness, transparency, and bandwidth but also identified motor specific force as a central scaling constraint. Direct drive is therefore an architecture with specific advantages and penalties, not a synonym for maximum efficiency or performance.[11]

Low-ratio or [quasi-direct drive](https://aiwiki.ai/wiki/quasi_direct_drive) architectures occupy a middle ground. The MIT Cheetah actuator study combined a torque-dense motor, low-ratio transmission, and careful thermal design to seek high torque density, force-control bandwidth, and impact backdrivability. Those properties resulted from the whole actuator and robot design, including motor geometry, gear ratio, sensing, structure, and control. Copying only the ratio does not reproduce the reported performance.[12]

Electric actuation is convenient when the available energy is electrical and when independent distributed joints are desired. It avoids centralized fluid lines, and regenerative drives can sometimes return mechanical energy to an electrical bus. Losses can still occur in batteries, cabling, inverters, windings, magnetic materials, bearings, and transmissions. Whether energy regeneration is useful depends on the motion, bus architecture, storage system, and drive controls. Holding torque can require continuous current unless a brake, latch, counterbalance, spring, or self-locking mechanism carries the load.[3][6]

## Hydraulic actuation

Hydraulic actuators use pressurized liquid to produce force or torque. A cylinder produces linear motion; a hydraulic motor produces rotation. The actuator belongs to a circuit that can include a prime mover, pump, reservoir, accumulator, valves, filters, heat exchanger, hoses or rigid lines, and sensors. A centralized hydraulic power unit can serve several actuators, which complicates comparisons with self-contained electric drives. Counting only a cylinder's mass or only a motor's mass excludes different parts of the required systems.[2][5]

Liquids have much lower compressibility than gases under ordinary operating conditions, but hydraulic systems are not perfectly rigid. Fluid compressibility, entrained air, hose expansion, valve dynamics, structural compliance, leakage, seal friction, and load inertia affect response. An accumulator can store pressurized fluid and provide peak flow, smooth pulsation, or maintain pressure, but it also stores hazardous energy. Servo-hydraulic systems can achieve precise controlled motion when their valves, sensors, mechanics, and controllers are designed for the task. Precision is not excluded by the use of hydraulics, just as it is not guaranteed by electricity.[2]

Hydraulics are often chosen when high force must be produced within a compact cylinder envelope or when shock-resistant machinery already has a hydraulic supply. They also introduce fluid cleanliness requirements, leakage paths, temperature-dependent viscosity, return-line and back-pressure effects, and maintenance of seals and hoses. A leak can reduce control authority and contaminate the environment. Design must consider relief paths, hose or tube ratings, load holding, safe depressurization, and the behavior following loss of power or a line failure. ISO 4413:2010 gives general rules and safety requirements for hydraulic fluid-power systems and components; it remained the published edition as of the research cutoff.[13]

Hydraulic efficiency cannot be represented by one number for the whole class. Pump type and operating point, pressure margin across metering valves, leakage, line losses, cooling, accumulator strategy, and duty cycle can dominate. A valve-controlled system that throttles a large pressure drop behaves differently from a displacement-controlled or electrohydrostatic actuator. The 2023 comparative experiment found the tested electric system drew less power, but its hydraulic rig had no accumulator and used a particular valve-controlled circuit. The authors explicitly bounded the setup, which is why the result should not be converted into a universal claim.[2]

## Pneumatic actuation

Pneumatic actuators use compressed gas, usually air. Cylinders, rotary vanes, air motors, bellows, and pneumatic muscles are common forms. A compressor, receiver, treatment equipment, regulators, valves, tubing, silencers, and sensors may be needed around the actuator. Gas compressibility gives the system appreciable energy storage and compliance. It can also make accurate intermediate positioning more difficult when compared with a mechanically stiff drive, unless suitable sensing, proportional valves, and control are used.[2][14]

For a conventional cylinder, ideal force begins with pressure times effective area, as in hydraulics. Actual force is reduced or altered by opposing chamber pressure, seal friction, flow restriction, and changing pressure during motion. Because compressed air expands, pressure at the cylinder can differ dynamically from the supply regulator setting. Tube volume, valve conductance, exhaust restriction, load, and piston velocity all matter. End-to-end motion between mechanical stops is a different task from continuously controlled position or force along the stroke.[5][14]

Pneumatic actuation is often useful for simple repetitive motions, compliant contact, rapid clamping, or facilities that already supply clean compressed air. Exhaust air can be released locally rather than returned through a fluid line, but generating and conditioning compressed air consumes energy and can produce noise. The absence of liquid does not make every pneumatic system clean enough for every environment, nor does compressibility make every pneumatic actuator intrinsically safe. Stored pressure, unexpected motion, hose failure, pinch points, and exhaust hazards remain relevant. ISO 4414:2010 supplies general rules and safety requirements for pneumatic systems and components.[14]

Pneumatic and hydraulic circuits can both contain stored energy after the command or main power is removed. A trapped volume, elevated load, accumulator, receiver, or compressed elastic element can continue to drive motion. Safe service requires isolation, controlled dissipation, and verification of a zero-energy state appropriate to the machine. The actuator's nominal energy source therefore does not by itself describe its fail-safe state.[13][14]

## Material and soft actuators

Some actuators generate motion through a material's deformation rather than a conventional motor, cylinder, or gearbox. Examples include shape-memory alloys, dielectric elastomers, ionic polymer-metal composites, piezoelectric stacks, magnetostrictive materials, electroactive polymers, and thermally expanding elements. These devices span very different physical effects and scales. Their useful output is determined by strain, stress, cycle rate, efficiency, voltage or temperature limits, fatigue, environmental sensitivity, packaging, and the mechanism that converts local deformation into task motion.[15][16]

A shape-memory alloy can recover a trained shape when a phase transformation is induced by temperature. Resistive heating is a common means of activation in robotic prototypes. Compact geometry and useful force are attractive, but temperature-dependent transformation, hysteresis, cooling time, fatigue, and nonlinear sensing and control limit the contexts in which it is appropriate. The phrase "artificial muscle" is descriptive, not a guarantee that the actuator reproduces biological muscle's efficiency, bandwidth, force-length behavior, or life.[17]

Dielectric elastomer actuators place a compliant dielectric between electrodes. Applied electric field produces electrostatic stress and deformation. They can provide large strain and lightweight compliant structures, but high electric fields, dielectric breakdown, electrode durability, viscoelasticity, environmental effects, fabrication variability, and lifetime are active engineering challenges. Reporting a material strain from a laboratory specimen is not the same as reporting blocked force, load-bearing stroke, usable bandwidth, or lifetime of a packaged robotic actuator.[15]

Fluidic elastomer actuators deform when internal chambers are pressurized. Channel geometry and reinforcement translate expansion into bending, twisting, extension, or contraction. Their bodies can conform to objects and distribute contact, which is useful in grippers and wearable devices. The required pump, compressor, valves, and tubing can remain rigid or bulky even when the actuator body is soft. Softness also complicates state estimation because a continuously deformable structure has more possible shapes than a rigid one-degree-of-freedom joint.[16]

A McKibben pneumatic artificial muscle consists of an internal bladder surrounded by a braided sleeve. Pressurization changes the braid geometry and typically causes axial contraction over its intended range. Force depends on pressure, length, braid angle, bladder and sleeve mechanics, and losses; hysteresis and pressure dynamics matter. Chou and Hannaford's experiments and model therefore characterize tension as a function of both pressure and length rather than assigning the device one constant force.[18]

The term [artificial muscle](https://aiwiki.ai/wiki/artificial_muscle) also covers actuator families that do not share the McKibben mechanism. Comparisons should separate the active material from the full actuation system and specify whether mass and power electronics, fluid supply, thermal management, sensors, and load-bearing structure are included. Reviews of soft actuators emphasize that promising laboratory performance does not remove the needs for reproducible manufacture, reliable interfaces, control, power delivery, and long cycle life in real applications.[15][16]

## Robotic joint architectures

A robotic actuator often integrates several components into a joint module: motor, reduction, bearings, housing, position sensor, drive electronics, thermal path, cable routing, and sometimes a brake or torque sensor. Integration reduces assembly burden and can make ratings easier to reproduce, but it also creates coupled limits. A compact housing can restrict heat rejection. A large hollow shaft can ease cable routing while reducing available material. A high-ratio gearbox can multiply torque while raising reflected inertia and friction. The module's output bearing may limit moment load even when its motor and reducer can supply the advertised torque.[3][6]

The familiar stiff geared servo favors position authority and compact torque multiplication. It is common in an [industrial robot](https://aiwiki.ai/wiki/industrial_robot), where repeatable motion, payload, working envelope, and process forces are evaluated at the robot level. Gear and bearing compliance still exist, and backlash, friction, controller gains, calibration, temperature, load, and pose affect endpoint behavior. NIST's robot-characterization work distinguishes accuracy from repeatability and notes that load, path, dynamics, compliance, and test conditions influence reported performance.[7]

A [series elastic actuator](https://aiwiki.ai/wiki/series_elastic_actuator) intentionally places a compliant element between the drive and load. If spring stiffness is known, measuring deflection can estimate transmitted force or torque. Pratt and Williamson argued that the spring can filter shock loads, reduce effective reflected inertia in interaction, and support stable force control, while accepting a tradeoff in zero-motion force bandwidth. A NIST physical-component library separately models rotational springs, dampers, gears, and inertias, illustrating why the elastic element and load-side dynamics must be explicit in a system model. Series elasticity is therefore a mechanical design choice, not merely a software setting.[19][20]

Variable-impedance actuators extend this idea by changing apparent or physical stiffness, damping, equilibrium position, or inertia. The literature distinguishes impedance shaped actively by feedback from mechanisms with inherent fixed or adjustable compliance and damping. A mechanism that adjusts spring geometry has different energy use, bandwidth, and failure behavior from a controller that emulates a spring using a stiff drive. Stiffness, damping, and inertia should be named separately rather than bundled under the vague adjective "compliant."[21]

[Tendon-driven actuation](https://aiwiki.ai/wiki/tendon_driven) places motors or other drives away from the joint and transmits tension through cables or tendons. Remote placement can reduce distal mass and permit compact joints, but tendons carry tension rather than compression and may require antagonistic routing, pretension, return springs, or gravity. Routing curvature, pulley radius, elasticity, friction, creep, and attachment compliance affect the relationship between motor motion and joint torque. Tendon-driven is a transmission architecture and can be combined with electrical, pneumatic, hydraulic, or material actuation.[16]

Legged and [humanoid robots](https://aiwiki.ai/wiki/humanoid_robot) place demanding constraints on joint actuation because repeated impacts, changing leverage, high peak power, balance, and limited onboard energy occur together. There is no single "humanoid actuator." Some systems use high-ratio geared electric joints, others use low-ratio drives, remote tendons, hydraulic cylinders, or elastic elements. Published legged-robot studies show that motor torque density, transmission ratio, impact backdrivability, thermal capacity, and force-control bandwidth must be designed together.[11][12]

Actuator arrangement also changes whole-body dynamics. Distal actuator mass increases limb inertia. Remote actuation can reduce that inertia but adds routing and compliance. Springs can return energy in cyclic motion but only when their stiffness and timing suit the trajectory. A transmission optimized for slow stance torque may constrain swing speed, while one optimized for speed may require more motor current at high load. These are task-specific tradeoffs rather than shortcomings of a named architecture.[11][12][19]

## Sensing and control

Position sensing may occur at the motor, transmission output, joint, cylinder rod, or external mechanism. A motor-side [rotary encoder](https://aiwiki.ai/wiki/rotary_encoder) measures rotor angle accurately but does not directly observe gearbox compliance, backlash, cable stretch, or load-side deflection. Dual encoders can separate motor and joint motion. A linear encoder can measure screw or cylinder output. Resolver, Hall sensor, potentiometer, inductive, optical, magnetic, and vision-based measurements each have environmental and bandwidth constraints.[4]

Force and torque can be sensed directly with strain-based transducers or estimated from pressure, spring deflection, motor current, or a dynamic model. Every estimate has a physical boundary. Motor-current torque estimation includes motor torque but must account for rotor acceleration and downstream losses. Cylinder pressure can estimate piston force only with correct areas and allowance for friction and dynamics. Spring-deflection sensing measures the force carried by that spring. A six-axis force-torque sensor at an end effector measures interaction at its mounting interface, not each joint's internal load.[5][6][20]

Position control reduces error between commanded and measured position. Velocity control regulates motion rate. Torque or force control regulates mechanical interaction. Impedance control commands a relationship between motion and force, while admittance control uses measured force to generate a motion command. These labels describe control laws and interfaces; they do not establish performance without sampling rate, filtering, sensing location, actuator dynamics, saturation, delays, and load conditions.[4][21]

Contact makes actuator dynamics especially important. High mechanical impedance can produce a large impact impulse before software responds. Passive compliance acts without sensor or computation delay, but it also stores energy and can introduce resonance. Low-ratio backdrivable drives reduce some reflected inertia, but motor and link inertia remain. Force-control benchmarks proposed by NIST include settling time, overshoot, steady-state error, response to an obstruction, transition between position and force modes, and disturbance handling. These metrics test the coupled robot and controller rather than a catalog torque alone.[22]

Artificial intelligence can supply higher-level commands, estimate state, adapt control parameters, or learn policies for [robot manipulation](https://aiwiki.ai/wiki/robot_manipulation) and [robot locomotion](https://aiwiki.ai/wiki/robot_locomotion). It does not change an actuator's physical power, temperature, pressure, stress, or speed limits. A learned controller must still respect saturation, latency, stability margins, collision constraints, and safe states. Simulation-to-real differences in friction, compliance, backlash, delays, and thermal behavior can be particularly consequential when a policy drives hardware near its limits.[4][22]

Learning can also support actuator diagnosis or calibration by modeling friction, estimating load, detecting anomalies, or compensating repeatable nonlinearities. Such compensation should not be confused with eliminating the underlying wear, leakage, insulation damage, fatigue, or structural defect. Safety-relevant monitoring needs specified coverage, independent validation, and an action when uncertainty or a fault exceeds its limits. A model output alone is not a protective mechanical stop, relief valve, brake, or energy-isolation device.[22][23]

## Selection and specification

A useful actuator specification starts with a load case and motion profile rather than a preferred technology. The designer identifies output coordinate, force or torque versus time, required speed and acceleration, range or stroke, duty cycle, reversals, impacts, external moments, and expected environment. The design then maps those requirements through the mechanism to the actuator. Static holding, accelerating a mass, overcoming friction, and absorbing an impact can impose different limits.[3][7]

| Criterion | Questions to define | Common reporting error |
| --- | --- | --- |
| Force or torque | Continuous, intermittent, peak, stall, or load-side? At what speed and temperature? | Treating peak or stall output as continuous |
| Speed and stroke | At what load, supply voltage or pressure, and transmission ratio? | Quoting no-load speed as loaded speed |
| Power and energy | Mechanical output or electrical/fluid input? Instantaneous or cycle average? | Omitting pump, compressor, drive, or stored energy |
| Precision | Accuracy, repeatability, resolution, backlash, or compliance? Under which path and load? | Using the terms interchangeably |
| Dynamic response | Mechanical resonance, closed-loop bandwidth, settling time, or impact response? | Assigning one bandwidth to an uncontrolled component |
| Thermal duty | Ambient, cooling, cycle, winding or fluid limit, and permitted derating? | Applying a short-duration rating indefinitely |
| Mass and volume | Bare transducer, joint module, or full supply system? | Comparing inconsistent boundaries |
| Safety and failure | Loss of power, sensor fault, broken line, jam, runaway, and stored energy? | Assuming power removal stops motion |

Force-density and torque-density figures require a declared denominator. Motor torque divided by motor mass excludes gearbox, bearings, housing, drive, cabling, and cooling. Joint torque divided by module mass is a different measure. A hydraulic cylinder can appear extremely compact if pump, reservoir, valves, and accumulator are excluded. A pneumatic muscle can appear light if compressor and valves are excluded. Either boundary can be valid, but it must be stated and applied consistently.[2][3]

Efficiency also depends on operating point. Electric motor efficiency varies with speed and torque; gearbox efficiency varies with load, ratio, direction, lubrication, and temperature. Hydraulic losses depend on pump and valve operation, pressure margin, leakage, and flow. Pneumatic energy use depends on compression, treatment, distribution, exhaust, and pressure. Material actuators may spend energy heating and then wait for passive cooling. A single best-case component efficiency cannot represent the energy per completed task.[2][6]

Accuracy is closeness to a commanded or true location, while repeatability is the spread when a motion is repeated under defined conditions. Resolution is the smallest distinguishable command or measurement increment, not proof of accuracy. Backlash is lost motion associated with direction reversal, while compliance is displacement under load. A mechanism can have high encoder resolution but poor absolute accuracy, low backlash but appreciable elastic deflection, or good repeatability after warm-up but a temperature-dependent offset. NIST's characterization framework treats these as separate performance measures.[7]

Bandwidth likewise needs a definition. Electrical current-loop bandwidth, motor mechanical bandwidth, transmission resonance, joint torque-control bandwidth, and robot endpoint response are not interchangeable. A compliant element may reduce impact force while lowering one resonance or control bandwidth. High gains can improve tracking until sensor noise, delay, compliance, saturation, or unmodeled dynamics cause instability. Published results should name the input, output, amplitude, operating point, and measurement method.[12][19][22]

Service life is governed by the limiting component and loading history. Bearings and rolling screws are often evaluated against load spectra; gears against tooth and flexural stresses; cables against bending and tension cycles; seals against pressure, speed, fluid, surface condition, and contamination; and polymers against strain, environment, and fatigue. Shock loads can dominate even when average torque is modest. A quoted cycle count from one component or laboratory test should not be assigned to every actuator built with the same technology.[9][15]

Maintainability includes access, lubrication, seal or cable replacement, calibration, contamination control, and safe isolation. Distributed electric joints may be individually replaceable but can require specialized electronics. Centralized hydraulics consolidate power generation but distribute fluid lines. Tendons can move motors to accessible locations but add routing and tension adjustment. Integrated modules simplify installation but may make internal repair impractical. The appropriate choice depends on the machine's service environment and downtime constraints.[3][13][14]

## Safety and reliability

An actuator creates or controls mechanical energy, so its hazards extend beyond normal motion. Crushing, shearing, entanglement, ejection, unexpected startup, dropped loads, hot surfaces, electrical shock, high-pressure injection, hose whip, stored elastic energy, and uncontrolled gravity motion can all be relevant. Risk assessment begins with the machine, intended use, foreseeable misuse, people exposed, and operating modes. ISO 12100:2010 specifies general principles for machinery risk assessment and risk reduction and remained the published edition as of the cutoff.[23]

Protective measures can include inherently safer geometry, force or speed limitation, guarding, interlocks, mechanical stops, brakes, counterbalances, pressure relief, load-holding valves, energy isolation, redundant sensing, monitored control, and validated stopping functions. The correct measure depends on the hazard. A torque limit inferred from motor current may not limit gravity motion after power loss. A normally closed brake can hold a joint but creates its own stopping, release, wear, and diagnostic requirements. A relief valve limits pressure at its sensing point but does not prevent all motion or hose failures.[13][23]

Industrial robot safety is addressed by ISO 10218-1:2025 for industrial robots and ISO 10218-2:2025 for industrial robot applications and cells. Part 1 treats the robot as partly completed machinery; Part 2 addresses integration into an application or cell. Their scopes exclude several domains, including medical, healthcare, public-access service, and consumer robots. The standards therefore should not be cited as universal requirements for every robot containing an actuator.[24][25]

ISO/TS 15066:2016 supplements the industrial-robot standards for collaborative industrial robot systems and work environments. It does not apply to non-industrial robots, although some safety principles may be informative elsewhere. ISO/PAS 5672:2023 specifies test methods for measuring transient and quasi-static forces and pressures in human-robot contacts in collaborative applications. A force-limited actuator is only one part of such an application; tool shape, robot mass, speed, control, workpiece, body region, and integration also affect contact risk.[26][27]

Different application domains use different safety frameworks. ISO 13482:2014 covers specified personal-care robot types and human-robot physical contact, while excluding industrial robots and robots that are medical devices. As of July 28, 2026, ISO listed it as published but expected to be replaced by a new edition. Medical rehabilitation and assistive robots may instead fall under medical-device requirements such as IEC 80601-2-78:2019, depending on the product and jurisdiction. A designer must establish the applicable scope rather than choose a standard because an actuator resembles one used in another robot.[28][29]

Reliability engineering examines both component failure and system response. Encoder loss, overtemperature, drive transistor failure, winding short, jammed gearbox, broken tendon, seal leakage, stuck valve, pressure loss, delamination, and fatigue crack lead to different consequences. Diagnostics may detect some faults, but safe behavior also depends on whether gravity, external force, or stored pressure can move the mechanism afterward. Failure-mode analysis should state assumptions and test the actual architecture.[13][14][23]

## Applications

Actuators appear in manufacturing equipment, process valves, vehicles, aerospace systems, construction machinery, prosthetic and rehabilitation devices, consumer products, and robots. The same task can often be performed by multiple technologies. A factory clamp may use a pneumatic cylinder, electric screw, hydraulic cylinder, toggle mechanism, or motor-driven cam. The choice reflects available utilities, required force and stroke, cycle rate, cleanliness, controllability, footprint, maintenance, and safety.[2][3]

In manipulators, actuators move each joint and the end effector. A high-payload arm may favor compact high-ratio joints or hydraulics; a force-sensitive assembly system may prioritize low friction, load-side sensing, and force-control response. In mobile robots, actuator mass and efficiency affect battery life and locomotion. In wearable systems, alignment, comfort, noise, mass distribution, compliance, and safe human contact can outweigh a raw peak-force figure. A review of wearable robotic orthoses consequently evaluates complete actuation systems across electric, hydraulic, pneumatic, and smart-material approaches rather than declaring one universal solution.[30]

Soft grippers and continuum robots exploit actuator deformation to conform to objects or environments. This can reduce reliance on exact geometric models, but it does not remove the need for sensing, pressure or voltage limits, fatigue testing, and task-specific control. Rigid and soft components are frequently combined: a soft chamber may be driven by rigid valves, a tendon by a conventional motor, or a compliant end effector by a stiff robot arm.[15][16]

In AI-enabled robots, actuators are the physical channel through which planned or learned actions affect the world. Better perception, [motion planning](https://aiwiki.ai/wiki/motion_planning), or [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) can improve how that channel is used, but software cannot make an actuator deliver force, energy, or bandwidth beyond its validated envelope. Reliable robotic behavior therefore joins data and algorithms with mechanics, power, sensing, real-time control, thermal design, and safety engineering.[4][22]

## See also

- [Robot](https://aiwiki.ai/wiki/robot)
- [Servo motor](https://aiwiki.ai/wiki/servo_motor)
- [Harmonic Drive](https://aiwiki.ai/wiki/harmonic_drive)
- [Cycloidal drive](https://aiwiki.ai/wiki/cycloidal_drive)
- [Leadscrew](https://aiwiki.ai/wiki/leadscrew)
- [Degrees of Freedom](https://aiwiki.ai/wiki/degrees_of_freedom)
- [Robot safety](https://aiwiki.ai/wiki/robot_safety)
- [Robot learning](https://aiwiki.ai/wiki/robot_learning)

## References

[1] International Organization for Standardization. "ISO 8373:2021 Robotics - Vocabulary." 2021. https://www.iso.org/standard/75539.html

[2] Jan Pustavrh, Marko Hocevar, Primoz Podrzaj, Ana Trajkovski, and Franc Majdic. "Comparison of hydraulic, pneumatic and electric linear actuation systems." Scientific Reports, 2023. https://pmc.ncbi.nlm.nih.gov/articles/PMC10684514/

[3] Kevin M. Lynch and Frank C. Park. "Modern Robotics: 8.9 Actuation, Gearing, and Friction." Northwestern University and Cambridge University Press. https://modernrobotics.northwestern.edu/nu-gm-book-resource/8-9-actuation-gearing-and-friction/

[4] Kevin M. Lynch and Frank C. Park. "Modern Robotics: 11.1 Control System Overview." Northwestern University and Cambridge University Press. https://modernrobotics.northwestern.edu/nu-gm-book-resource/11-1-control-system-overview/

[5] Parker Hannifin Corporation. "Designing With Cylinders." Mobile Cylinder Division engineering guide. https://www.parker.com/parkerimages/mobilecylinder/cat/english/0001q.pdf

[6] maxon. "Motor Data and Operating Ranges." Engineering reference. https://www.maxongroup.com/medias/sys_master/8798985748510.pdf?attachment=true

[7] Nicholas Dagalakis. "Robot Characterization Testing." NISTIR 4510, National Institute of Standards and Technology, 1991. https://nvlpubs.nist.gov/nistpubs/Legacy/IR/nistir4510.pdf

[8] John E. Huber, Norman A. Fleck, and Michael F. Ashby. "The selection of mechanical actuators based on performance indices." Proceedings of the Royal Society A, 1997. https://doi.org/10.1098/rspa.1997.0117

[9] THK. "Ball Screw General Catalog: Selection and preload." Manufacturer engineering reference. https://tech.thk.com/en/products/pdf_download.php?file=E_15_BallScrew.pdf

[10] Harmonic Drive LLC. "Harmonic Drive General Catalog: Strain Wave Gearing." Manufacturer engineering reference. https://www.harmonicdrive.net/_hd/content/catalogs/pdf/general_catalog.pdf

[11] Gavin Kenneally, Avik De, and Daniel E. Koditschek. "Design Principles for a Family of Direct-Drive Legged Robots." IEEE Robotics and Automation Letters, 2016. https://kodlab.seas.upenn.edu/uploads/Gavin/gake_dddesign.pdf

[12] Patrick M. Wensing, Albert Wang, Sangok Seok, David Otten, Jeffrey Lang, and Sangbae Kim. "Proprioceptive Actuator Design in the MIT Cheetah: Impact Mitigation and High-Bandwidth Physical Interaction." IEEE Transactions on Robotics, 2017. https://fab.cba.mit.edu/classes/865.18/motion/papers/mit-cheetah-actuator.pdf


[13] International Organization for Standardization. "ISO 4413:2010 Hydraulic fluid power - General rules and safety requirements for systems and their components." https://www.iso.org/standard/44781.html

[14] International Organization for Standardization. "ISO 4414:2010 Pneumatic fluid power - General rules and safety requirements for systems and their components." https://www.iso.org/standard/44790.html

[15] Meng Li, Aniket Pal, Amirreza Aghakhani, Abdon Pena-Francesch, and Metin Sitti. "Soft actuators for real-world applications." Nature Reviews Materials, 2022. https://pmc.ncbi.nlm.nih.gov/articles/PMC7612659/

[16] Daniela Rus and Michael T. Tolley. "Design, fabrication and control of soft robots." Nature, 2015. https://dspace.mit.edu/bitstream/handle/1721.1/100772/SoftRoboticsReview-FinalAuthorVersion.pdf

[17] Deivamoney Josephine Selvarani Ruth, Jung-Woo Sohn, Kaliaperumal Dhanalakshmi, and Seung-Bok Choi. "Control Aspects of Shape Memory Alloys in Robotics Applications: A Review over the Last Decade." Sensors, 2022. https://pmc.ncbi.nlm.nih.gov/articles/PMC9269604/

[18] Ching-Ping Chou and Blake Hannaford. "Measurement and Modeling of McKibben Pneumatic Artificial Muscles." IEEE Transactions on Robotics and Automation, 1996. https://bdml.stanford.edu/twiki/pub/HSR/HumanSafeRobot/ChouHannaford96.pdf

[19] Gill A. Pratt and Matthew M. Williamson. "Series Elastic Actuators." Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, 1995. https://fab.cba.mit.edu/classes/865.15/people/rebecca.kleinberger/assets/papers/SEA_Pratt.pdf

[20] Charles A. Manion, Conrad Bock, and Raphael Barbau. "Physical Component Libraries for SysPhS Modeling and Simulation in Manufacturing." NISTIR 8490, National Institute of Standards and Technology, 2023. https://nvlpubs.nist.gov/nistpubs/ir/2023/NIST.IR.8490.pdf

[21] Bram Vanderborght et al. "Variable Impedance Actuators: A Review." Robotics and Autonomous Systems, 2013. https://ris.utwente.nl/ws/portalfiles/portal/6520436/pp.pdf

[22] Joe Falco, Jeremy Marvel, Rick Norcross, and Karl Van Wyk. "Benchmarking Robot Force Control Capabilities: Experimental Results." NISTIR 8097, National Institute of Standards and Technology, 2016. https://nvlpubs.nist.gov/nistpubs/ir/2015/NIST.IR.8097.pdf

[23] International Organization for Standardization. "ISO 12100:2010 Safety of machinery - General principles for design - Risk assessment and risk reduction." https://www.iso.org/standard/51528.html

[24] International Organization for Standardization. "ISO 10218-1:2025 Robotics - Safety requirements - Part 1: Industrial robots." https://www.iso.org/standard/73933.html

[25] International Organization for Standardization. "ISO 10218-2:2025 Robotics - Safety requirements - Part 2: Industrial robot applications and robot cells." https://www.iso.org/standard/73934.html

[26] International Organization for Standardization. "ISO/TS 15066:2016 Robots and robotic devices - Collaborative robots." https://www.iso.org/standard/62996.html

[27] International Organization for Standardization. "ISO/PAS 5672:2023 Robotics - Collaborative applications - Test methods for measuring forces and pressures in human-robot contacts." https://www.iso.org/standard/82488.html

[28] International Organization for Standardization. "ISO 13482:2014 Robots and robotic devices - Safety requirements for personal care robots." https://www.iso.org/standard/53820.html

[29] International Electrotechnical Commission. "IEC 80601-2-78:2019 Medical electrical equipment - Part 2-78: Particular requirements for basic safety and essential performance of medical robots for rehabilitation, assessment, compensation or alleviation." https://webstore.iec.ch/en/publication/33594

[30] Allan Joshua Veale and Sheng Quan Xie. "Towards compliant and wearable robotic orthoses: A review of current and emerging actuator technologies." Medical Engineering & Physics, 2016. https://research.utwente.nl/en/publications/towards-compliant-and-wearable-robotic-orthoses-a-review-of-curre/