robotics//actuator

An actuator is a device that converts a control signal into physical work (torque, displacement, pressure, heat, light or flow), and it is what lets a controller act on the world instead of only observing it. It is the outward half of a robot or a plant: a sensor measures the world, the actuator changes it. Motors, solenoids, hydraulic cylinders, valves and heaters are all actuators, and the useful classification is by energy domain and failure mode rather than by shape. It is the link most often forgotten on the whiteboard and the one that rules in practice: the controller produces a number, and everything between that number and the force it was meant to be belongs to the actuator.

An actuator is a device that converts a control signal into physical work (torque, displacement, pressure, heat, light or flow), and it is what lets a controller act on the world instead of only observing it. It is the outward half of a robot or a plant: a sensor measures the world, the actuator changes it. Motors, solenoids, hydraulic cylinders, valves and heaters are all actuators, and the useful classification is by energy domain and failure mode rather than by shape. It is the link most often forgotten on the whiteboard and the one that rules in practice: the controller produces a number, and everything between that number and the force it was meant to be belongs to the actuator.

A first model needs only two facts, a limit and a lag. Whatever the controller asks, the actuator delivers at most its maximum (saturation), and it takes time to get there: a drone's motor and propeller need 10 to 100 ms to change speed, a pneumatic process valve seconds to travel its stroke. Written down, the delivered value is ureal=sat⁡(ucmd)u_{real}=\operatorname{sat}(u_{cmd})ureal​=sat(ucmd​) and the output follows it with a time constant, τmω˙=−ω+k ureal\tau_m\dot\omega=-\omega+k,u_{real}τm​ω˙=−ω+kureal​. When the actuator is much faster than the process (a valve on a tank that takes minutes to fill) it can be treated as a pure gain; its dynamics and its limit enter the design when they are comparable to the loop, or when the loop asks a lot of it.

Five properties describe any actuator, and a drone and a factory show each one. Saturation: a maximum thrust, and a minimum that is positive because a fixed-pitch propeller cannot pull down; a valve between 0 and 100 %. Dynamics: tens of milliseconds for a motor, seconds for a valve. A dead zone with friction: below idle a motor does not respond, and a valve does not move until friction is overcome, then jumps (stiction). Nonlinearity: thrust grows with the square of motor speed, flow follows the valve's characteristic (control valve). Quantization: about 2,000 throttle levels in a digital ESC protocol, a finite resolution in a valve positioner.

The command is not the motion. Backlash in gears makes the output lag every reversal of the input, and with dead zones and dry friction it produces the small limit cycles that make a loop hunt. Those ordinary nonlinearities and the actuator's dynamics, more than any tuning, determine what a controller can physically achieve (control).

Guarantees end at the actuator's authority. A safety filter that keeps a robot out of a wall by braking, or a control barrier function, promises nothing once the braking it demands exceeds what the motors can give; every such guarantee carries the actuator limits as an assumption.

It wears, and wear changes the loop. A valve's packing loosens or tightens, a motor's bearings wear (brushless motor), and a loop that is nervous moves its actuator more and wears it faster; the command, and not only the output, is worth watching.

An actuator plus a sensor closes the loop. Without feedback, software is issuing wishes to physics.