Optical Feedback Control

• Optical feedback control is a framework where optical fields act as both sensors and actuators to modify system dynamics in applications like stabilization, cooling, and phase tuning.
• It encompasses measurement-based, adaptive, and coherent feedback methods applied in systems ranging from semiconductor lasers to quantum optomechanics.
• Its implementations in devices such as interferometers, spatial light modulators, and integrated photonics enable precise noise reduction, delay management, and mode control.

Searching arXiv for recent and foundational papers on optical feedback control to ground the article in published work. Optical feedback control denotes a class of control architectures in which optical fields are both the sensing and actuation medium, so that light modified by a plant is either measured and converted into a classical control signal or coherently recirculated to influence the plant dynamics directly. Across quantum optics, optomechanics, semiconductor lasers, integrated photonics, gravitational-wave interferometry, and beam-shaping systems, the central controlled effect is the modification of an optical or optomechanical system’s effective dynamics through delayed, phase-sensitive, or measurement-conditioned reinjection of optical information. The topic spans measurement-based feedback, adaptive measurement, and coherent feedback; targets include stabilization, cooling, phase estimation, entanglement enhancement, frequency tuning, suppression of instability, and control of nonlinear or stochastic spatiotemporal dynamics (Serafini, 2012, Ernzer et al., 2022, Pan et al., 26 Jun 2026).

1. Foundational architectures and control paradigms

Three broad paradigms recur across the literature. In measurement-based feedback, an optical output is measured, processed electronically or digitally, and then returned as a control action. This structure appears in homodyne-controlled optomechanics, Kalman-filtered levitated nanoparticles, adaptive phase estimation, and cavity-optomechanical entanglement control (Berni et al., 2015, Kremer et al., 2023, Asjad et al., 2016). In adaptive measurement, the measurement basis itself is modified in real time, as in homodyne phase estimation where the local-oscillator phase is shifted toward the Fisher-information optimum (Berni et al., 2015). In coherent feedback, the output field is not measured at all, but coherently transformed and fed back as a quantum optical signal, so that both field quadratures remain available inside the loop (Hamerly et al., 2012, Ernzer et al., 2022).

A second organizing distinction concerns the role of delay and phase. In semiconductor and cavity-based systems, the return phase of a delayed field determines whether feedback is constructive, destructive, stabilizing, or destabilizing. This is explicit in dual-wavelength lasers, VCSEL polarization feedback, DFB lasers on silicon photonics, and optomechanical coherent-feedback cooling (Pawlus et al., 2022, Wang et al., 3 Feb 2026, Hauck et al., 2016, Ernzer et al., 2022). In spatially extended optical media, the feedback can be effectively instantaneous in time but nonlocal in space, as when a blurred optical image is fed back to an optically addressed spatial light modulator (Semenov et al., 2023).

A further distinction separates feedback that primarily modifies a plant’s dissipation from feedback that primarily modifies its effective potential, spectral response, or state-estimation accuracy. The provided sources emphasize linewidth control, damping engineering, mode competition, spatial front control, and output-state shaping rather than a single universal feedback law. This suggests that “optical feedback control” is better treated as a family of architectures than as one canonical protocol.

2. Linear quantum-optical and optomechanical feedback

In linear quantum-optical settings, optical feedback control is often formulated in state-space, Langevin, or input-output form. For a levitated nanoparticle modeled as a linear optomechanical system, the dynamics are written as a linear state-space equation,

$\mathbf{\dot{z}(t) = \mathbf{A}\mathbf{z} + \mathbf{B}{u}(t) + \mathbf{w}(t) + \mathbf{u}_{op}(t),$

with a noisy continuous measurement

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$

Kalman state estimation, and an LQG-type control law

$u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$

that minimizes a quadratic cost (Kremer et al., 2023). In that work, the controlled degree of freedom is the nanoparticle’s axial motion, the actuation enters through the mechanical momentum equation, and the dynamics are explicitly linear and Gaussian. The same source states that there are no nonlinear mechanical terms such as $x^3$ or $x^4$, and no explicit delay terms, so its contribution to optical feedback control is a measurement-estimation-actuation architecture rather than nonlinear potential engineering (Kremer et al., 2023).

A related but more explicitly quantum-noise-engineering framework appears in cavity optomechanics with feedback-controlled in-loop light. There, balanced homodyne detection of transmitted light is used to drive an acousto-optic modulator that modulates the cavity input field, so that the optical input itself becomes a feedback-shaped quantum resource (Bemani et al., 2021). The feedback law is

$$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$

and the loop renormalizes the cavity linewidth to

$$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$

The same analysis shows that the effective in-loop optical input acquires thermal-like and anomalous correlations, and that the feedback-modified cavity equation contains a $c^\dagger$ term formally analogous to intracavity parametric amplification (Bemani et al., 2021). In the optimal low-noise regime, the system is therefore analogous to an optomechanical system containing a near quantum-limited optical parametric amplifier coupled to an engineered reservoir interacting with the cavity (Bemani et al., 2021).

Coherent all-optical feedback furnishes a third linear framework. For open quantum oscillators in the LQG setting, coherent controllers are represented as optical cavities, OPOs, or more general linear quantum systems embedded in an interferometric loop (Hamerly et al., 2012). The state-space description is constrained by physical realizability conditions,

$$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$

and the steady-state covariance is obtained from

$A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$

That work shows that coherent control schemes can outperform optimal measurement-based feedback control schemes in the quantum regime of low steady-state excitation number, attributing the advantage to the coherent controller’s ability to simultaneously process both quadratures of an optical probe field without measurement or loss of fidelity (Hamerly et al., 2012).

3. Phase, delay, and interference as control resources

In many optical-feedback systems, control is exerted primarily through phase-sensitive delayed self-interaction. A paradigmatic integrated-photonics example is dual-wavelength laser control by a common external feedback cavity designed so that the two wavelengths accumulate a relative phase shift of approximately $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$0: $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$1 With an electro-optic phase modulator tuning the common feedback phase, one wavelength can be resonant while the other is anti-resonant, so the feedback effectively boosts one mode and penalizes the other (Pawlus et al., 2022). The corresponding two-mode Lang–Kobayashi-type model is

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$2

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$3

with coupled carrier dynamics. Experimentally, that single phase actuator yielded extinction ratios of up to $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$4 dB for a $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$5 nm separation and up to $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$6 dB for a $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$7 nm separation (Pawlus et al., 2022).

A closely related but stabilization-oriented example is a commercial DFB laser edge-coupled to a silicon photonics PIC with a tunable optical reflector based on a ring-resonator add-drop multiplexer (Hauck et al., 2016). The laser frequency shift is modeled by

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$8

with normalized feedback strengths

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$9

The paper states that the equations are guaranteed to have only one solution for $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$0, while larger effective feedback can produce multiple external-cavity modes and bifurcation (Hauck et al., 2016). Experimentally, phase variation via laser-chip spacing led to sawtooth-like frequency shifts, linewidth broadening from about $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$1 MHz to about $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$2 MHz near bifurcation, and up to $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$3 additional modes in unstable phase windows. At lower effective feedback, ring tuning provided few-GHz mode-hop-free frequency tuning, with measured full tuning ranges such as $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$4 GHz and $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$5 GHz depending on feedback phase (Hauck et al., 2016).

A semiconductor-laser example with explicitly nonlinear delayed dynamics is the VCSEL with polarization-engineered optical feedback (Wang et al., 3 Feb 2026). There the external cavity length is $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$6 m, corresponding to a roundtrip delay of about $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$7 ns, and the RF spectra show peaks separated by $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$8 MHz, exactly the inverse of the cavity round-trip time (Wang et al., 3 Feb 2026). The control parameter is the half-wave-plate angle $u(t) = -\mathbf{k^T} \mathbf{\hat{z}}$9, which changes the polarization state and effective power of the reinjected field, thereby regulating the nonlinear interaction between TE and TM modes. The TM mode then exhibits heavy-tailed fluctuations and extreme events defined by

$x^3$0

At $x^3$1 mA, the number of extreme TM events in a $x^3$2s record changed non-monotonically with angle, reaching $x^3$3 at $x^3$4, $x^3$5 at $x^3$6, and $x^3$7 at $x^3$8, with corresponding maximum intensities $x^3$9 mV, $x^4$0 mV, and $x^4$1 mV (Wang et al., 3 Feb 2026). This suggests that optical feedback control is not restricted to stabilization; it can also be used for controlled destabilization and rare-event engineering.

4. Feedback cooling, damping engineering, and quantum control of mechanics

Cooling and damping engineering constitute one of the most developed branches of optical feedback control. A survey of quantum-optical feedback identifies early amplitude-squeezing enhancement, QND memories, feedback cooling, adaptive measurements, and coherent-feedback strategies as central application areas (Serafini, 2012). Within that landscape, optomechanical feedback provides especially clear examples of how optical measurements or optical recirculation reshape mechanical susceptibility.

For a membrane-in-the-middle force sensor with feedback-controlled in-loop light, the mechanical response and optically added noise are expressed as

$x^4$2

and in the resolved-sideband RWA the on-resonance response becomes

$x^4$3

The same work reports that in a symmetric cavity at $x^4$4 and unit efficiency, $x^4$5, whereas in a strongly asymmetric cavity with $x^4$6, $x^4$7, allowing it to become much smaller than the SQL threshold $x^4$8. It also reports maximum response $x^4$9 for a symmetric cavity and $$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$0 for a strongly asymmetric cavity (Bemani et al., 2021). The physical picture is that feedback reduces the effective cavity linewidth and creates phase-sensitive optical correlations, thereby enabling gain without the usual gain-bandwidth tradeoff.

Coherent optical feedback can act even more directly on mechanical damping and backaction. In an experimental platform where a light field interacts twice with the same membrane mode through two different cavity modes, the simplified unresolved-sideband, on-resonance feedback-induced damping and spring shift are

$$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$1

This directly shows that delay sets whether the feedback is primarily momentum-like or position-like, while optical phase controls the sign and magnitude of the force (Ernzer et al., 2022). The same work derives a minimum phonon number bound

$$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$2

achieved for $$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$3 and $$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$4, and experimentally reports cooling a membrane mode to

$$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$5

phonons, corresponding to $$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$6, in a $$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$7 K environment (Ernzer et al., 2022). The paper states that this lies below the theoretical limit of cavity dynamical backaction cooling in the unresolved sideband regime and is achieved with only $$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$8 of the optical power required for cavity cooling (Ernzer et al., 2022).

Measurement-based cold damping can also be redirected toward communication objectives rather than lowest temperature alone. In a two-mode optomechanical cavity, a derivative feedback force based on homodyne measurement of the phase quadrature of one output mode takes the form

$$c_{\rm in}^{(1)}(t) = c_{\rm in,0}^{(1)}(t) + \Phi (t), \qquad \Phi(t)= g\, i_{\rm fb}(t-\tau),$$9

and the feedback-modified effective damping at $$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$0, $$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$1 simplifies to

$$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$2

The heating term is canceled when

$$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$3

That feedback-enhanced regime increases output entanglement and can push coherent-state teleportation fidelity above the secure threshold $$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$4 even after loss, while also enabling two-way Gaussian steering in parameter regimes where this would not occur without feedback (Asjad et al., 2016).

5. Spatial, thermal, and many-body optical feedback control

Not all optical feedback loops are most naturally described as mode-by-mode linear control. In beam-shaping and spatially extended optics, the controlled variable may be a beam profile or a domain pattern rather than a single quadrature.

A macroscopic and technologically motivated example is feedback control of optical beam spatial profiles using thermal lensing (Liu et al., 2013). A four-segmented heater around an SF57 transmissive optic acts as a multifunctional optical actuator capable of spherical lensing, cylindrical lensing, and beam steering. The local linearized plant is represented by a $$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$5 transfer matrix,

$$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$6

and the feedback controller uses the inverse of this transfer matrix together with slow bias adjustment and local PI control (Liu et al., 2013). In a symmetric aberration experiment, all four measured beam parameters returned to their set points within approximately $$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$7 s after each disturbance. The work also reports how beam-radius and beam-pointing fluctuations increase when the aberrator is heated strongly in air, attributing the degradation mainly to convection and noting that the intended application is in vacuum (Liu et al., 2013).

In an optically addressed spatial light modulator subject to optical feedback, the feedback is spatial and nonlocal rather than pointwise. The governing model includes a Gaussian-convolved feedback field and a local retardation dynamics

$$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$8

with the blue intensity determined by interference between the incident and feedback fields (Semenov et al., 2023). By tuning $$\kappa_{\rm fb} = \kappa - g\sqrt {2\eta\kappa _1\kappa _2}.$$9 so that the local dynamics behaves like pitchfork or saddle-node normal forms, the authors show deterministic control of front direction and coarsening speed. They also introduce colored noise in the green illumination,

$c^\dagger$0

and report that around $c^\dagger$1, noise can invert front propagation in one operating regime (Semenov et al., 2023). This suggests that optical feedback control can act on effective asymmetry and front motion rather than merely on stability in the narrow sense.

Many-body quantum optical feedback forms a further extension. For ultracold atoms in an optical lattice inside a cavity, the conditional master equation with feedback is

$c^\dagger$2

with jump operator $c^\dagger$3 and feedback superoperator $c^\dagger$4 in the instantaneous limit (Mazzucchi et al., 2016). In the weak-measurement regime, the gain parameter $c^\dagger$5 acts like an effective time offset for coherent tunneling, leading to a stabilization law

$c^\dagger$6

for a target collective observable, and a critical gain

$c^\dagger$7

For bosons and fermions, the paper reports stabilization of density-wave imbalance and antiferromagnetic order, respectively, with qualitatively different regimes for $c^\dagger$8 and $c^\dagger$9 (Mazzucchi et al., 2016). This broadens optical feedback control from few-mode or single-device regulation to trajectory-level steering of correlated many-body states.

6. Applications, performance regimes, and emerging control frontiers

Application domains now range from metrology to large-scale interferometry. In adaptive optical phase estimation with squeezed light, a rough Bayesian estimate is first used to compute a correction

$$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$0

which is then applied to the local oscillator so that the final homodyne stage occurs near the Fisher-information optimum (Berni et al., 2015). With measured squeezing $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$1 dB and anti-squeezing $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$2 dB, and an experimentally optimal phase $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$3, the protocol achieved deterministic phase estimation below the coherent-state shot-noise benchmark and below the heterodyne limit (Berni et al., 2015). Here the feedback loop optimizes information extraction rather than directly stabilizing a plant state.

In levitated optomechanics, a recent adaptive LQG architecture stabilizes a nanoparticle at the unstable intensity minimum of an optical double-well potential (Mlynář et al., 14 Aug 2025). The reduced local dynamics around the apex is

$$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$4

with LQR objective

$$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$5

and indirect adaptive tracking of the drifting apex via an augmented Kalman-Bucy filter (Mlynář et al., 14 Aug 2025). In nonlinear simulations with realistic timing, the adaptive $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$6D controller reduced the standard deviation of the tracking/confined $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$7-measurement signal to about $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$8 mV, versus about $$A\Theta + \Theta A^{\rm T} + B J B^{\rm T} = 0, \qquad \Theta C^{\rm T} + B J D^{\rm T} = 0, \qquad D J D^{\rm T} = J,$$9 mV for the non-adaptive $A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$0D controller and about $A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$1 mV for the adaptive $A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$2D controller (Mlynář et al., 14 Aug 2025). This suggests that optical feedback control is increasingly being used not only to cool around stable equilibria but to stabilize deliberately anti-restoring operating points that are useful for low-absorption quantum experiments.

At the scale of gravitational-wave detectors, optical feedback has recently been demonstrated as a suppression mechanism for optomechanical parametric instability in Advanced LIGO (Pan et al., 26 Jun 2026). The parametric gain is given by

$A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$3

and the instability criterion is $A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$4. The control strategy senses the beat note between the carrier TEM$A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$5 mode and the PI-involved TEM$A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$6 higher-order mode at the OMC photodetector, uses that beat note as an error signal, and drives an AOM to generate a TEM$A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$7-profiled sideband that is converted by cavity mode mismatch into a TEM$A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$8 control field (Pan et al., 26 Jun 2026). Experimentally, the unstable $A\sigma + \sigma A^{\rm T} + B F B^{\rm T} = 0.$9 kHz mode was suppressed from uncontrolled $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$00 to a controlled value

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$01

consistent with the abstract statement $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$02 (Pan et al., 26 Jun 2026). This is an example of optical feedback controlling an instability by directly canceling the optical higher-order mode that mediates the optomechanical loop, rather than damping each mechanical mode individually.

A different application frontier is programmable microwave photonics based on microcombs (Moss, 2024). There, the governing transversal filter is

$y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$03

and the control objective is accurate realization of the tap weights $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$04 through iterative optical or impulse-response feedback (Moss, 2024). The paper compares four feedback methods and reports that average deviation $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$05 is achieved after about $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$06 iterations for one-stage methods and about $y(t) = \mathbf{C}\mathbf{z}(t)+y_n(t),$07 iterations for two-stage methods in the reported examples, with the two-stage synergic spectral-plus-impulse method performing best in both temporal integration and RF filtering tasks (Moss, 2024). This suggests that optical feedback control is becoming an enabling calibration layer for reconfigurable photonic processors rather than merely a stabilization accessory.

Across these regimes, a recurrent practical lesson is that optical feedback control is constrained by loss, delay, detector efficiency, and nonlinear actuator physics, but gains power from exactly those optical features that are liabilities in open loop: phase accumulation, interference, cavity selectivity, and amplitude-phase coupling. A plausible implication is that future optical-feedback systems will increasingly combine optical-domain actuation, state estimation, and integrated monitoring rather than relying on a single pure paradigm. The literature surveyed here already spans autonomous bath-mediated self-locking in quantum dots (Ladd et al., 2010), coherent double-pass quantum control (Ernzer et al., 2022), measurement-conditioned many-body stabilization (Mazzucchi et al., 2016), and full-scale interferometer instability suppression (Pan et al., 26 Jun 2026), indicating that optical feedback control has become a unifying methodology across widely separated photonic platforms.