Feedback

When you reach for a glass of water, you don't just fire off your arm muscles once and hope. Your eyes track the glass continuously. Your brain notices if your hand is drifting slightly left and corrects the trajectory. Your fingers feel when they've made contact and adjust grip force so you don't crush the glass or drop it. The entire motion is a constant loop of doing and adjusting, doing and adjusting — guided by information flowing back from your senses.

That loop is feedback, and it is one of the most powerful ideas in all of engineering.

In control systems — the field that underpins how robots are commanded — feedback means using a measurement of a system's output to adjust its input. Instead of computing the "correct" command once and sending it, a feedback controller continuously measures what is actually happening and closes the gap between where the system is and where it should be.

Open-loop versus closed-loop control

A system without feedback is called open-loop. A timer that runs a sprinkler for ten minutes regardless of soil moisture is open-loop. A robot arm that moves to a preset angle without checking whether it arrived is open-loop. Open-loop control is simple and cheap but fragile: any disturbance — a gust of wind, a heavier-than-expected load, a slight miscalibration — causes error that accumulates uncorrected.

A system with feedback is called closed-loop. The output is measured (using a sensor), compared to the desired output (the setpoint), and the difference (the error) is used to generate a corrective action. The loop is "closed" because information flows from output back to input, completing a circuit.

PID: the workhorse of feedback control

The most widely deployed feedback controller in the world is the PID controller — standing for Proportional, Integral, Derivative. It computes a corrective command from three terms: a term proportional to the current error (react to how far off you are now), a term based on the accumulated error over time (correct persistent biases), and a term based on how fast the error is changing (dampen oscillations). Tuning a PID controller — setting the weights for each term — is a fundamental skill in robotics engineering. Virtually every motor controller, temperature regulator, and flight-stabilisation system in the world uses some variant of PID.

Why feedback is irreplaceable

Robots operate in a physical world that is noisy, variable, and full of surprises. Loads change, surfaces shift, batteries drain, motors wear. A robot controlled without feedback is a robot that assumes the world will cooperate perfectly — an assumption that fails constantly. Feedback is what makes control robust: it doesn't matter exactly why the error arose, only that it exists and must be reduced. This property — correcting for disturbances without needing to know their cause — is what makes feedback controllers so durable in practice.

If feedback is so powerful, why do some highly capable robots deliberately limit how much feedback they use — and what does that reveal about the hidden costs of always knowing where you are?