Focus mode · Press Esc to exit

Module 1 · Lesson 1 · 12 min video · 20 min practice

Feedforward, feedback, and disturbances

See why a plan alone cannot reject surprises, and why measurement closes the loop.

  • Distinguish feedforward from feedback using signal flow.
  • Identify reference, output, error, control input, and disturbance.
  • Predict which architecture rejects an unmeasured disturbance.
Unofficial TUM ACE alignment: ACE §1: Regelungsaufgabe und Wirkungsplan

Visual lesson storyboard ready · video recording next

8–15 minute visual explanation. The interactive experiment below is usable now.

Predict before the explanation

A drone knows the exact motor command that holds it at two metres in still air. A gust arrives. What happens?

Commit to an answer: does the fixed command recover the height, or does the drone need information about what actually happened?

A plan is useful—until reality deviates from the plan

Feedforward computes an action from what we want and what we know about the plant. If the model and disturbance are exact, it can be fast and elegant. But an unmeasured gust is absent from that calculation.

Feedback measures the actual output, compares it with the reference, and acts on the error. It does not need to know whether the error came from wind, mass uncertainty, or a small motor mismatch. It only needs the consequence to be observable in the measurement.

Concept check

Which signal gives a controller evidence that an unmeasured gust changed the drone's height?

Reactive experiment: let the gust arrive

Move the disturbance. The feedforward command stays fixed; the feedback path responds to the measured error. This control and the result below are cells in one connected Python program.

The same slider now updates a result farther down the lesson—there is no run-all button and no iframe boundary.

Worked example: cruise control on a hill

At constant throttle, a car slows when the road tilts upward. A speed sensor exposes the drop. The controller increases throttle until the error is small again. Notice the division of labor:

  1. The reference says what speed is wanted.
  2. The sensor reports what speed occurred.
  3. The comparator forms the error.
  4. The controller converts error into throttle.
  5. The plant—car plus road—turns throttle and disturbance into speed.

Derivation

Why loop gain attenuates an output disturbance

For plant P, controller C, and an additive output disturbance d, the closed-loop equations are u = C(r-y) and y = Pu+d. Substitution gives y = PC(r-y)+d, hence (1+PC)y = PCr+d.

The disturbance-to-output transfer is therefore 1/(1+PC). Later we will name this function the sensitivity.

Unofficial TUM ACE exam prep

Name the signals in a Wirkungsplan

Practise translating between the international terms used here and the course notation: Führungsgröße, Regelabweichung, Stellgröße, Regelgröße, and Störgröße. The diagrams and questions on this site are original.

Guided exercise

Sketch two signal-flow diagrams for room heating: one with outdoor-temperature feedforward and one with indoor-temperature feedback. Then answer:

  • Which architecture can act before the room cools?
  • Which architecture can correct an unknown open window?
  • Why is combining them often better than choosing only one?
Hint

Feedforward is strongest when the disturbance is measurable; feedback is strongest when its effect appears in the controlled output.

Summary

  • Feedforward acts from a model and known inputs.
  • Feedback acts from the measured consequence of uncertainty.
  • Feedback trades corrective authority for sensitivity to noise, delay, and actuator limits—topics we will quantify later.

Retrieve it later

Without looking, draw the five signals in a feedback loop and show where an output disturbance enters.

Answer from memory tomorrow, then return to check your reasoning.