Sense, then correct
Picture driving down a straight road. You do not turn the wheel to one exact angle, lock your arms, and hope. You glance at the lane, notice you are drifting left, nudge right, glance again, nudge back. You are running a tiny loop dozens of times a second: look at where you are, compare it to where you want to be, and make a small correction. The road can curve, the car can pull to one side, and you still stay centered — because you keep checking the result of your own steering.
A thermostat does the same thing. It does not run the heater for a fixed number of minutes; it measures the room temperature, compares it to the number you set, and switches the heat on or off accordingly. So does your hand when you pour water to a line on a glass: you watch the level rise and ease off as you get close. This pattern — sense, compare, act, repeat — is the single most important idea in robot control, and everything else in this track is built on top of it.
The loop that runs many times a second
When an engineer turns that habit into something a machine does on its own, they call it feedback control: the robot feeds the result of its action back into its own decision about what to do next. Instead of a one-way command — "motor, turn to here" — the action and the measurement form a circle. The output of the system loops back around to influence the next input. That circle is the feedback loop, and naming it is the whole point of this guide.
The crucial detail is how fast it spins. A human steering a car closes the loop maybe a few times a second. A robot joint typically closes it hundreds or even thousands of times a second. Each pass is the same three beats — measure where you are, work out how far off you are, push to close the gap — but they happen so quickly that the motion looks smooth and continuous to us. The robot is really just inching toward its goal in tiny, constant corrections.
Four pieces every loop has
Pull any feedback loop apart and you find the same four pieces. Once you can name them, you can read almost any control diagram in robotics, because they are always these four wired in a circle.
- The set-point: where you want to be. This is the target — the lane center, the room temperature you dialed in, the angle you want a joint to reach. See set-point and error signal for the pair that drives the loop.
- The measurement: where you actually are, read by a sensor. For a robot joint this is usually a rotary encoder reporting the true angle. No measurement, no feedback.
- The error: set-point minus measurement — exactly how far off you are, and in which direction. A positive error might mean "you are short of the target"; a negative one, "you overshot."
- The corrective action: the command the controller sends to the actuator — say, the voltage to a DC motor — chosen to shrink the error. Bigger error, harder push; near zero error, ease off.
Notice that the error is the heart of the loop. Everything the controller does is a reaction to that one number. The simplest useful rule is just "push in proportion to the error," and the most common controller in all of robotics — the one you will meet next — is built almost entirely out of clever ways to react to the error and to how it changes over time.
Why this beats a fixed command
To feel why the loop matters, imagine a motor that must hold an arm level. The blind approach is to send one fixed command: "apply exactly enough power to hold the arm horizontal," then walk away. This is open-loop control — a command issued once with no checking. On a perfect day with the calibration just right, it works. The trouble is that the world is never that clean.
Now lean a small weight on the arm, or let the motor warm up so it pushes a little weaker, or just accept that your guess of "exactly enough power" was slightly off. The fixed command cannot know any of this, so the arm sags and stays sagging. A blind, one-shot order has no way to notice it is wrong, let alone fix it. Anything that pushes the system off target — an added load, friction, a breeze, a manufacturing quirk — is called a disturbance, and open-loop control is helpless against all of them.
Close the loop and the story changes completely. The instant the arm sags, the encoder reports a lower angle, the error grows, and the controller automatically pushes harder until the arm comes back to level — then holds it there, pushing exactly as hard as the disturbance demands, moment to moment. It never needed to know the weight, the temperature, or the friction in advance. It only needed to watch the error and react. That contrast is the subject of open-loop vs closed-loop control, and it is why nearly every real robot runs closed-loop.
The backbone of everything ahead
Step back and the loop is strikingly simple: a set-point, a measurement, an error, and an action, wired in a circle that spins many times a second. That same circle steers a self-driving car between the lines, keeps a drone hovering against a gust, and holds a surgical tool steady against the tremor of a human hand. The pieces get more sophisticated, but the shape never changes.
Everything else in this track is a deeper look at one part of this loop. How exactly should the action depend on the error? That gives you PID control and its three terms. How do you tune the reaction so it is firm but does not wobble? That is gain and tuning and stability. What happens when the motor cannot push any harder than it already is? That is saturation and wind-up. Hold the picture of the loop in your head, and each of those will land as a refinement of something you already understand.