Open- and Closed-Loop Control

The same command, a different result

A small robot is told to drive for two seconds and stop at a green line. On a smooth floor it nearly succeeds. On carpet it stops short; after a gentle push it travels sideways. The command did not change, but the world did.

Predict: What information would let the robot correct while it is still moving?

Where this fits in your learning path

In the previous module, you learned to observe, record, and replay robot evidence. This lesson turns that evidence path into an action path: a controller uses a current measurement—not merely a log saved for later—to revise what the robot does next.

Follow this three-lesson progression:

1. This lesson: decide whether outcome information closes the loop and compute signed error.
2. Next: use proportional control to turn error into command, then reason about overshoot and stability.
3. Then: combine proportional, integral, and derivative terms and tune them from evidence.

You do not need calculus or ROS 2 code here. Keep asking one question: does the measured outcome change the current command?

Two information structures

An open-loop controller sends an action without using a measurement of the controlled outcome to revise that action. A timer-driven motor command is open loop with respect to final position.

A closed-loop controller measures an output, compares it with a desired value, and uses the difference to update its command. The loop repeats while the task runs.

The sign matters. If the measurement is 1.07 m, then e = 1.00 - 1.07 = -0.07 m; the robot has passed the target.

Worked comparison

Open-loop attempt:

1. Command 40% motor effort for 2.0 s.
2. A low battery reduces acceleration.
3. The robot stops at 0.78 m.
4. Nothing in the controller detects the 0.22 m error.

Closed-loop attempt:

1. Set the target to 1.00 m.
2. Measure position repeatedly.
3. Compute error = target - measurement.
4. Update the command until position and motion are within defined tolerances.

Feedback can reject some disturbances and model errors because their effects appear in the measurement. It cannot correct a failure that is not observed, arrives too late, saturates the actuator, or makes the measurement misleading.

Open loop is not automatically bad

Open loop can be appropriate when the environment is predictable, the action is brief, variation is harmless, measurement is unavailable, or a trusted lower-level controller already closes the relevant loop. A camera exposure pulse and a prevalidated gripper release may be open loop at one layer while motor current is closed loop underneath.

Closed loop is not automatically safe

A feedback loop depends on its sensor, estimator, timing, controller, actuator limits, and physical plant. Noise can make the command chatter. Delay can make correction arrive late. Excessive gain can cause overshoot or oscillation. A biased sensor can drive the robot confidently toward the wrong value.

Control structure is a claim about information flow—not a guarantee of accuracy or safety.

Retrieval practice

For each task, identify the controlled variable, setpoint, measurement, disturbance, and one unobserved failure:

• a mobile robot stopping at a docking marker;
• a robot arm holding a joint angle;
• a fan running for five seconds to cool electronics.

Summary

• Open loop does not use measured outcome to revise the current action.
• Closed loop compares a setpoint with a measurement and acts on error.
• Feedback helps only when the relevant effect is observable and timely.
• One robot can contain nested open- and closed-loop decisions.
• Evaluate the entire sensing–control–actuation path, not the label alone.

Sources and further study

• MIT Underactuated Robotics — trajectory feedback — explains why open-loop trajectories can diverge and feedback can regulate toward a plan; retrieved 2026-08-20.
• MIT Underactuated Robotics — output feedback — formalizes decisions from noisy measured outputs; retrieved 2026-08-20.
• WPILib PIDController — documents setpoint error and tolerance checks in a widely used robotics control library; retrieved 2026-08-20.
• WPILib flywheel tuning — demonstrates feedback and combined feedforward-feedback response to disturbance; retrieved 2026-08-20.