Closed-loop control

A cyclist riding with their eyes open and a cyclist riding blindfolded are doing very different things. Eyes open, they see the kerb approaching, steer away from it, check they've moved enough, and correct again — thousands of micro-adjustments per second. Eyes closed, they pedal and hope. The cyclist with eyes open is running a closed-loop system. The one who is blindfolded is running open-loop.

The difference is feedback: information that flows back from the world into the decision-making process.

The feedback loop explained

A closed-loop control system has four parts working in a cycle:

1. Reference — the desired state (target speed, target angle, target temperature).
2. Sensor — a device measuring the actual current state.
3. Controller — compares reference against actual, computes the error, decides how to correct.
4. Actuator — applies the correction (motor, valve, heater).

Then the sensor measures again, the controller compares again, and the cycle repeats — dozens to thousands of times per second. Because the loop closes back on itself, any disturbance that pushes the system off target gets detected and corrected automatically.

A concrete example: a robotic arm joint

A robotic arm from KUKA or Fanuc needs to move its elbow joint to exactly 47 degrees. An encoder — a sensor that counts tiny notches on a rotating disc — reports the current angle 1,000 times per second. If the joint reaches only 44 degrees (perhaps because a heavy payload resisted), the controller detects the 3-degree error and applies more motor current. When the encoder reports 47 degrees, the controller reduces current to hold position. If the arm is nudged by a worker, the system corrects back within milliseconds. The arm doesn't "know" about the disturbance; it simply keeps closing the gap.

Why closed-loop control changed robotics

Open-loop systems are fine until something unexpected happens. Manufacturing, surgery, and autonomous driving are full of unexpected things: varying loads, temperature changes, wear, uneven terrain. Closed-loop control makes a robot robust to those surprises. A surgical robot using closed-loop force control can detect the resistance of tissue and stop before cutting too deep. A self-driving car using closed-loop steering can hold a lane even as crosswinds push it sideways.

The cost is complexity: you need sensors, and sensors can fail. You also have to tune the controller carefully — too aggressive a correction and the system oscillates; too timid and it responds slowly. This is precisely what PID controllers and more advanced methods like model-predictive control are designed to handle.

Almost everything precise in modern engineering — from a phone's image stabiliser to a nuclear plant's cooling system — runs a closed feedback loop of some kind.