Feed-forward

Feed-forward is a term describing a kind of system which reacts to changes in its environment, usually to maintain some desired state of the system. A system which exhibits feed-forward behavior responds to a disturbance without having to be affected by it first — contrast with a feedback system.

Feedback control is reactive; feed-forward control is pro-active. Feed-forward control can respond more quickly to known kinds of disturbances, but cannot do much with novel disturbances. Feed-back control deals with any deviation from desired system behavior, but requires the system to respond to the disturbance in order to notice the deviation.

Feed-back control is exemplified by homeostatic regulation of heartbeat in response to physical exertion. Feed-forward control can be likened to learned responses to known cues. These systems could be in control theory, physiology or computing.

Elements common to all feed-forward systems

A feed-forward system can be illustrated by comparing it with a familiar feedback system — that of cruise control in a car. When in use, the cruise control enables a car to maintain a steady road speed. When an uphill stretch of road is encountered, the car slows down below the set speed; this speed error causes the engine throttle to be opened further, bringing the car back to its original speed (a PI or PID controller would do this. Note that a good PID control will return the car to the original speed, after an initial transient response).

A feed-forward system on the other hand would in some way 'predict' the slowing down of the car. For example it could measure the slope of the road and, upon encountering a hill, would open up the throttle by a certain amount, anticipating the extra load. The car does not have to slow down at all for the correction to come into play.

However, other factors than the slope of the hill and the throttle setting influence the speed of the car: air temperature, pressure, fuel composition, wind speed, etc. Just setting the throttle based on a function of the slope may not result in constant speed being maintained. Since there is no comparison between the output variable, speed, and the input variable, it is not possible to resolve this problem with purely feed-forward control.

Fortunately, the two types of control are not mutually exclusive; the feed-forward system just described could be combined with the feed-back system of conventional cruise control to allow quick response with the feedback system cleaning up for any error in the predetermined adjustment made by the feed-forward system.

Feed-forward does not have the stability problems that feed-back can have. Feed-forward needs to be pre-calibrated cause → effect, feed-back does not. That is another way of saying what was said above - that feed-forward control applies to disturbances with known effects.

Physiological feed-forward system

Main article: Feed-forward (physiology)

In physiology, (also called a feed-forward homeostatic control system) is a homeostatic control system in which, the anticipatory effect that one intermediate exerts on another intermediate further along in the pathway allows the system to anticipate changes in a regulated variable.

Feed-forward systems in computing

Main article: Perceptron

In computing, feed-forward normally refers to a multi-layer perceptron network in which the outputs from all neurons go to following but not preceding layers, so there are no feedback loops.