Two-degree-of-freedom control: give tracking its own path
A separate reference path changes how the loop follows a target while leaving its response to a disturbance unchanged.
THE SHORT ANSWER
Two-degree-of-freedom (2DOF) control gives the reference its own path into the controller, separate from the feedback that reacts to the measured output. With u = kₚ(βr − y) + Fr, the reference weight β and feedforward gain F change only the reference response; the closed-loop pole and disturbance response depend on feedback alone. Static feedforward relies on the model and does not remove error caused by model mismatch.
Calculate what reference feedforward changes, and what it leaves to feedback.
A robust feedback loop may still miss its target. A separate reference path can change target tracking without retuning the feedback gain.
Review the previous lesson: Robustness: will your controller survive a different motor?First, picture it
Feedback reacts to measured output. A separate reference path can also prepare a command when the target changes. This gives two ways to shape behavior: one for how the loop reacts to output, another for how it receives the target.
Build the idea step by step
This bench uses proportional feedback, reference weight beta, and a separate reference feedforward gain F. It has no integrator. Changing F changes the response to a target, while keeping the nominal feedback pole and additive input-disturbance path unchanged. For a plant with static gain one and proportional gain two, F=0 leaves a unit-step error of one-third. Choosing F=1 removes that nominal constant-target error. The next example shows the calculation; model error can spoil the result.
Read the equations
The lab uses proportional feedback, a reference weight, and static reference feedforward. There is no integral term in this bench.
For the lab's first-order plant with time constant 2 seconds and static gain K, feedforward changes the reference numerator, not the closed-loop pole. Here R(s) is the transformed reference.
What the symbols mean
- reference, output, and applied input in the normalized model
- proportional feedback gain
- dimensionless reference weight
- static reference feedforward gain
- actual plant static gain
Setting s = 0 gives the final value for a unit step, K(kₚβ+F)/(1+Kkₚ). With K = 1, kₚ = 2, β = 1 and F = 1 it equals 1; with K = 0.7 it is 0.875.
Remove nominal tracking error with feedforward
Use plant gain 1, proportional gain 2, reference weight 1, and a unit reference step. Start with feedforward gain zero.
Set feedforward gain to 1, the inverse of the nominal plant gain. The final output becomes 1. The pole remains minus 1.5 inverse seconds; tracking is not instantaneous.
Keep this controller but lower actual plant gain to 0.7. The final output is now 0.875. Model-based feedforward does not automatically correct an unknown model error.
More detail and model assumptions
Compare an increment, not two totals
Substitution gives denominator 2s+1+Kkₚ in both paths. β and F occur only in the reference numerator. Therefore the separate disturbance-induced increment is unchanged when only β or F changes. At the default K=1, kₚ=2, d=−0.5, its limit is −1/6. The total output also contains the reference transient, so compare the dashed disturbance component rather than subtracting unmatched timestamps.
Connect the example to the lab
Keep feedback gain fixed. Change reference feedforward and compare target tracking with the separate disturbance response.
Open this experimentFeedforward uses a model; it does not measure accumulated error like an integrator. In this P-only bench, changing beta also changes final tracking. A stable unsaturated two-degree-of-freedom PI controller can instead use integral action to restore constant-target accuracy. Do not confuse the two implementations.
Check your understanding+
Does feedforward F=1 keep the final output at 1 when actual plant gain falls from 1 to 0.7?
No. With proportional gain 2 and reference weight 1, the final output is 0.875. Static feedforward corrects the nominal model; it does not accumulate and remove the remaining error.Solve these before opening the solutions
Work each problem on paper first. Open the answer to check yours, then the full solution if you need the steps, and mark how it went.
The lab's plant is 2ẏ = −y + K(u + d) with K = 1, and the controller is u = kp(βr − y) + Fr. With kp = 3 and β = 1, find the closed-loop pole and the final output for a unit reference step, first with F = 0 and then with F = 1.
Show answer
AnswerThe pole is −2 s⁻¹ in both cases. F = 0 gives a final output of 0.75; F = 1 gives 1.
Show full solution
Solution
The pole comes from the denominator, which contains no F.
Final values from Y/R at s = 0.
With F = 1 the feedforward alone supplies the input u = 1 that holds y = 1, so feedback no longer needs a steady error to supply it. The pole did not move.
Keep K = 1 and kp = 3, but set the reference weight to β = 0.5. Find the F that gives zero nominal steady error. Then compare the command u(0⁺) right after a unit reference step with that of β = 1, F = 1. What do you conclude about β and F in this P-only bench?
Show answer
AnswerF = 2.5. Both settings give u(0⁺) = 4 and the same Y/R = 4/(2s + 4): in this bench the reference response depends only on the sum kpβ + F.
Show full solution
Solution
Zero steady error.
Right after the step the output is still 0.
β and F enter Y/R only through kpβ + F, so any pair with the same sum gives the same reference response. They would differ in a controller with dynamics in the reference path, such as PI with setpoint weighting.
The disturbance path Y/D = 1/(2s + 4) contains neither β nor F.
Exam level. The controller kp = 3, β = 1, F = 1 was designed for K = 1, but the real plant gain is K = 1.25. A unit reference step is followed later by a constant input disturbance d = −0.4. Find (a) the closed-loop pole, (b) the final output from the reference alone, (c) the steady output change caused by the disturbance and (d) the total final output. Compare with the nominal plant.
Show answer
Answer(a) −2.375 s⁻¹. (b) 1.053, so the final value exceeds the target by 5.3%. (c) −0.105. (d) 0.947. For K = 1 the same quantities would be −2, 1, −0.1 and 0.9.
Show full solution
Solution
Closed-loop pole with the real gain.
Reference contribution.
Disturbance contribution.
Total final output.
The feedforward was exact only for K = 1; with K = 1.25 it overshoots the target in steady state. The disturbance increment depends on K and kp but not on F. Both errors call for integral action in the feedback, not a different F.
Mark each exercise after you check your answer. Marks stay in this browser.
Terms this lecture teaches
- two-degree-of-freedom control (2-DOF)2자유도 제어
- tracking (reference following)목표 추종
- reference (setpoint)목표값
- feedforward피드포워드
- disturbance rejection외란 억제
Frequently asked questions
What is the difference between feedforward and feedback?
Feedback acts on the measured output and can correct effects the model did not predict. Feedforward acts on the reference, or on a measured disturbance, through a model before any error appears, so it is fast but only as accurate as that model.
Does reference feedforward change the disturbance response?
No. Both transfer functions share the denominator 2s + 1 + Kkₚ, and β and F appear only in the reference numerator. With K = 1, kₚ = 2 and an input disturbance of −0.5, the disturbance-induced output change tends to −1/6 whatever F is.
Can feedforward replace integral action?
No. Feedforward uses the nominal model and does not accumulate error. With F = 1 the final output is 1 when the plant gain is 1, but 0.875 when the actual plant gain drops to 0.7; a two-degree-of-freedom PI controller would use its integral to restore constant-target accuracy.
Why is it called two-degree-of-freedom control?
Because two separate design choices shape the loop: the feedback path, which sets the closed-loop poles and disturbance rejection, and the reference path, which sets how the target is introduced. A one-degree-of-freedom controller acts only on the error r − y, so the two cannot be tuned independently.
Why the next lesson follows
So far we wrote continuous-time control laws. A digital controller updates at separate instants; its clock becomes part of the design.
Further reading
Finished the reading and the exercises? Mark the lecture as done to tick it on the syllabus. Progress stays in this browser.
EIGENROOM · https://eigenroom.com/theory/two-dof · Updated 2026-11-15