Repetitive Control¶

Feed error from a modeled recurrence period back into an operating controller to track periodic references or reject periodic disturbances.

Version: v1 · 2026-10-07 · History
Domain-specific #: 13998
Domain group: Applied Sciences & Engineering
Origin domain: Engineering & Design (beyond software)
Subdomain: Control Engineering → Engineering & Design (beyond software)
Aliases: Repetitive controller

Core Idea¶

Repetitive control is a continuing feedback-control method that retains error across a modeled recurrence period and uses corresponding earlier-cycle information to correct later plant input. It is used when a reference trajectory repeats or a disturbance has a repeatable periodic component. A conventional feedback loop reacts to current error; the repetitive branch adds a memory of the relevant period. Hara and colleagues analyze the period-delay internal model and its conditional tracking result.^[1]

The word repetitive describes the controller's period-linked error path, not a promise of perfect repetition or exact rejection in every implementation. A correctly configured, stable ideal internal model can force error toward zero for a modeled periodic class under its theorem's assumptions. Filtering, plant dynamics, stability limits, and a wrong or changing period can leave residual error. An attempted repetitive controller remains identifiable by its architecture even when it does not meet a performance target.^[1]

Structural Signature¶

Signature: controlled plant and measured output + recurring target or disturbance period + measured error + period-linked memory and return path → later-cycle correction; performance depends on plant and loop conditions.

Continuing controlled plant and measured output. The output of a running process is available for comparison and later actuation. Removing this loop leaves a periodic signal or stored trace, not a feedback controller.^[1]^[2]
Modeled recurrence period. A specified period determines which earlier error is brought back to a corresponding phase. The controller may be mistuned to the actual signal, but without any modeled recurrence horizon it is no longer the named period-memory architecture.^[1]
Tracking or rejection error. The difference between desired and measured behavior supplies the quantity to be retained. A reference-following robot and a voltage-regulating inverter form different errors while sharing this role.^[3]^[2]
Period-linked internal memory and return path. A delay or stored sequence passes earlier-cycle error into current correction. Remove this path and ordinary feedback can remain, but the repetitive control mechanism has gone.^[1]^[2]
Stability and implementation shaping. Plant dynamics, loop gain, compensation, and optional filters determine whether that correction improves tracking or destabilizes the loop. Stability is necessary for the cited accuracy claims, not for recognizing a configured controller. A particular filter, notch, digital memory size, or advance is case-specific.^[1]^[2]
Tracking or disturbance-response readout. Compare the relevant output against the relevant reference or disturbance condition before claiming error reduction. The architecture can be identified without a successful test, but no measured accuracy follows from its name alone.^[3]^[2]

What It Is Not¶

A periodic reference is not itself repetitive control. A plant may receive a repeating command without returning prior-cycle error. Ordinary instantaneous feedback can also track a periodic command imperfectly; the missing discriminator is the retained period-linked correction.^[1]

Nor is the method identical to all iterative learning control. Hara contrasts his continuously operating repetitive loop with the specific betterment methods he cites, which restart separate trials from a common initial condition. That historical comparison does not classify every later learning-control design. A fixed period-memory feedback law also need not change an agent's durable capability, so the everyday word “learning” is insufficient to make the Learning Prime a parent.^[1]

Scope of Application¶

The method's literal habitat is control of plants exposed to recurring commands or disturbances. Omata and colleagues' manipulator follows a periodic trajectory with nonlinear compensation and position and velocity feedback. Zhang and colleagues' UPS inverter uses a digital repetitive branch to reduce periodic output-voltage distortion under nonlinear loads. The robot case concerns reference tracking; the inverter case concerns power-output regulation and disturbance rejection. In both, prior-period error affects later control action.^[3]^[2]

The accessible Omata original is a publisher abstract. It warrants the named method, periodic trajectory, three-link experiment, and reported low tracking error, but not its exact memory block, numeric error, or tuning coefficients. The Zhang original exposes its particular memory and compensator design. These two carriers establish a shared method within control engineering; they do not establish a generic memory mechanism in unrelated domains.^[3]^[2]

Clarity¶

The identity separates three easily conflated claims. Architecture: is there a continuing feedback loop with period-linked error memory? Match: does the actual command or disturbance recur at the modeled period? Result: is the implemented closed loop stable and accurate under a measured test? An affirmative answer to the first question does not imply the other two. Hara's ideal period-delay generator explains why a harmonically matched signal could be suppressed, while his modified designs show why real implementations need separate stability and tracking judgments.^[1]

It also distinguishes a repeating reference from a repeating disturbance. The manipulator is asked to repeat a motion; the inverter must preserve its output despite cyclic distortion from a nonlinear load. Both use period-linked correction, but they do not measure the same quantity or guarantee the same result.^[3]^[2]

Manages Complexity¶

A controller design can involve plant dynamics, sampling, phase, filters, compensation, resonance, and load behavior. The abstraction reduces recognition to four checks: what output is measured, what period is modeled, what error is retained, and how does it re-enter later input? Only after those checks should one examine loop stability and case-specific readouts. This keeps Zhang's notch and low-pass filter in the role of implementation choices rather than quietly making them necessary to every repetitive controller.^[1]^[2]

The compression has a limit. It does not turn a poor period estimate into a valid zero-error theorem, and it does not supply one universal recurrence equation. Hara develops ideal and modified forms; Zhang's digital DSP implementation is one realization. Preserve the period, plant, and filter assumptions when moving from a control diagram to a performance claim.^[1]^[2]

Abstract Reasoning¶

Suppose voltage distortion persists after a repetitive branch is enabled. First verify that the memory horizon corresponds to the fundamental output cycle, then check whether measured voltage error actually returns through the branch to later inverter input. If it does, the architecture is present. Next inspect the actual disturbance spectrum, loop stability, and filter or notch choices before inferring why the distortion remains. The original inverter study gives one worked case of such an implementation and reported output measurements; it does not make its percentages portable to every UPS.^[2]

For a periodic robot motion, ask whether the recurring command produces a phase-corresponding tracking error that can be reused while the plant continues operating. If the controller only corrects within the present cycle, ordinary feedback may explain the response. If it stores error across periods, repetitive control is the better classification. Omata's accessible abstract supports the applied classification, while Hara supplies the general period-memory mechanism; the exact robot memory realization remains uninspected.^[3]^[1]

Knowledge Transfer¶

Inside control engineering, the role test transfers from recurring robot trajectories to inverter output regulation: each new plant requires its own output sensor, recurrence period, error definition, return path, stability analysis, and accuracy measure. The robot's nonlinear compensation cannot simply be copied into the inverter, nor can the inverter's digital notch and THD figure be transferred into robot control.^[3]^[2]

The broader Feedback Prime travels farther because a measured output can return to later input without a modeled period. A repeating signal may instantiate Periodicity even with no controller. This entry's specialist method requires their particular control-engineering combination; word resemblance to spaced repetition or human learning is analogy, not an established instance.^[1]

Examples¶

Canonical: periodic trajectory tracking on a three-link manipulator¶

Omata and colleagues describe nonlinear repetitive control for a manipulator commanded along a periodic trajectory. Their original publisher abstract reports nonlinear compensation, position and velocity feedback, and a three-link experimental result with low tracking error. It does not expose numerical errors or the exact memory parameters. The period-linked mechanism below is therefore a method-level mapping from the named scheme and Hara's account, rather than a claim that those hidden robot blocks were inspected.^[3]^[1]

Mapped back: the continuing controlled plant and measured output are the manipulator and its position and velocity observations; the modeled recurrence period is the repeated trajectory command; the tracking error compares actual and commanded motion; the period-linked internal memory and return path are the repetitive controller named by Omata, whose exact implementation is not visible in the abstract; stability and implementation shaping include their stated nonlinear compensation, with full analysis uninspected; the readout is the reported low tracking error, with no numerical magnitude asserted.^[3]

Applied: SPWM UPS inverter under nonlinear load¶

Zhang and colleagues install a direct repetitive controller in an SPWM inverter supplying UPS power. The fundamental output cycle sets the digital memory horizon; the bridge-rectifier RC load is a source of output distortion, not the timing clock. In their tested design, a DSP stores period-linked integrated error, while a low-pass element and zero-phase notch shape the return path. They report output-voltage THD of 1.4–1.7% under tested nonlinear loads and error settling after about three to five fundamental periods when the repetitive branch is activated. Those figures are results for their apparatus.^[2]

Mapped back: the continuing controlled plant and measured output are the inverter and sensed output voltage; the modeled recurrence period is the fundamental output cycle; the rejection error is voltage-reference discrepancy that includes recurring distortion; the period-linked internal memory and return path are the DSP-stored integrated error and correction branch; stability and implementation shaping use the study's notch, low-pass element, and compensator; the readout is case-specific THD and settling. Remove the repetitive branch and the same SPWM inverter remains, but without period-memory correction.^[2]

Structural Tensions¶

T1: ideal periodic accuracy versus stable implementation. Hara's ideal period-delay internal model places resonant poles at modeled harmonics, which supports a conditional zero-error proposition for the matched periodic class. A real plant must also remain stable under feedback and high-frequency effects. Optional low-pass shaping can relax difficult stability conditions but leave residual high-frequency tracking error. Leaning toward ideal gain without the required plant conditions risks an unstable loop; leaning toward robust filtering limits attainable rejection. Diagnostic: for the plant and disturbance actually at hand, which frequencies need correction, and what stability and residual-error bounds does the chosen filter support?^[1]

This tradeoff describes design choices, not a fixed performance frontier common to every installation. Zhang's filters are one case response; Omata's abstract does not reveal enough detail to place its robot on the same quantified curve.^[2]^[3]

Structural–Framed Character¶

The entry lies toward the structural side of the structural–framed spectrum within its engineering habitat. Evaluative weight: a smaller error is a design objective, but the architecture can be recognized independently of whether a test succeeds. Human-practice dependence: engineers choose the reference, recurrence period, sensor, and acceptable error; the feedback relation itself operates in the plant once configured. Institutional origin: control engineering supplies the design language and test conventions, but no single institution is a constitutive component of the loop. Vocabulary travel: “repetition” and “learning” have broader uses; the named method travels only when prior-period measured error changes later plant input. Import versus recognition: labeling a learning exercise or periodically scheduled action as repetitive control would import an engineering loop absent from the case. The more portable skeleton is Feedback, the independently accepted Prime, while this period-memory specialization remains within the evidenced controller cases. Its character: a largely structural control method whose validity and performance are framed by chosen periods, plants, measurements, and stability conditions.^[1]^[3]^[2]

Structural Core vs. Domain Accent¶

The skeletal relation is Feedback: an identifiable plant output is measured and returned to change later input through a closed cause-and-effect path. The domain accent is a controller's modeled recurrence period and retained corresponding-phase error, which specialize that loop for periodic tracking or rejection. Hara's ideal delay and Zhang's digital memory are different implementations of this specialist addition.^[1]^[2]

The named entry does not clear the Prime bar on these sources. Both unlike positive cases remain engineered feedback controllers; no evidence here establishes a period-error controller across independent noncontrol substrates. Feedback has that broader parent role without the period-memory condition. Periodicity describes the target signal's recurrence, but a mismatched controller remains an instance of the architecture even if the actual disturbance does not repeat as expected. Learning would require a durable change in an agent's capability that the fixed feedback law does not need.^[1]

This entry is a kind of Feedback.

The strict Feedback parent is recorded in the typed DAG edge: each admitted repetitive controller closes a measured output-to-later-input loop, and its period memory is the narrower differentia. Periodicity is related because the reference or disturbance motivates a period model, but it is not a direct controller parent: the modeled period may mismatch the actual signal. Learning is a historical analogy in Hara's terminology, not a necessary agent-capability change.^[1]

A PID controller is another feedback implementation; it need not store corresponding-phase error from an earlier period. Iterative learning control is a neighboring family, and Hara's specific betterment comparison uses reset trials; that comparison is insufficient to turn every later iterative-learning design into the contrary of continuous control.^[1]

Relationships to Other Abstractions¶

Current abstraction Repetitive Control Domain-specific

Parents (1) — more general patterns this builds on

Repetitive Control is a kind of Feedback Prime

A measured error returns through period-linked memory to change later plant input.

Hierarchy path (1) — routes to 1 parentless root

Repetitive Control → Feedback

Neighborhood in Abstraction Space¶

Repetitive Control sits in a sparse region of the domain-specific corpus (94^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2551 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Periodicity: a repeating property of a signal. Repetition can exist without a feedback plant or stored error.
Ordinary feedback or PID control: measured error changes input, but no period-linked memory is required by those broader identities.
Reset-based betterment experiments: the specific historical comparison repeats trials from a common start, whereas Hara's repetitive loop continues across periods.^[1]
Ideal internal model theorem: the period-delay generator explains a conditional tracking result; it does not guarantee that any filtered installed controller has zero error.^[1]
A successful harmonic-rejection result: measured THD or tracking improvement is a case outcome, not the identity of the architecture.^[2]

References¶

[1] Shinji Hara, Yutaka Yamamoto, Tohru Omata, and Michio Nakano, “Repetitive Control System, A New Type Servo System for Periodic Exogenous Signals” (title punctuation transcribed as a comma for the reference binder; the original prints a colon), IEEE Transactions on Automatic Control 33, no. 7 (1988): 659–668, https://doi.org/10.1109/9.1274. Full original at Kyoto University, https://repository.kulib.kyoto-u.ac.jp/server/api/core/bitstreams/09b05ef9-9e66-4326-9cba-6d3bc017d8ec/content; especially printed pp.659–664 (PDF pp.1–6). registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v

[2] Kai Zhang, Yong Kang, Jian Xiong, and Jian Chen, “Direct Repetitive Control of SPWM Inverter for UPS Purpose,” IEEE Transactions on Power Electronics 18, no. 3 (2003): 784–792, https://doi.org/10.1109/TPEL.2003.810846. Original author-uploaded full article at https://www.researchgate.net/publication/3280382_Direct_repetitive_control_of_SPWM_inverter_for_UPS_purpose; especially printed pp.786–790, §§III–V. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r

[3] Tohru Omata, Shinji Hara, and Michio Nakano, “Nonlinear Repetitive Control with Application to Trajectory Control of Manipulators,” Journal of Robotic Systems 4, no. 5 (1987): 631–652, https://doi.org/10.1002/rob.4620040505. Original publisher abstract at https://onlinelibrary.wiley.com/doi/abs/10.1002/rob.4620040505; full article not inspected. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k