Cloak: Advances in Theory and Applications

• Cloak is a broad term describing devices, media, and algorithms that conceal physical or digital signatures across waves, diffusion, and computational systems.
• Transformation cloaks use mathematical mappings to redirect wave trajectories and control phase restoration, enabling partial invisibility in various fields.
• Cloaking in computing and security employs adversarial techniques and trusted execution environments to hide process identities and protect access patterns.

Searching arXiv for the specified "cloak" papers and closely related uses of the term. Search 1: general cloak query on arXiv. “Cloak” denotes a class of devices, media, and algorithms that suppress, redirect, or falsify an observable signature under a specified observation model. In wave physics, a cloak aims to make electromagnetic, acoustic, thermal, or matter-wave fields outside a protected region behave as if the region were absent or differently shaped; in transport and diffusion problems, the criterion is that trajectories, gradients, or asymptotic flow remain unchanged; in computing, “cloaking” refers to hiding process identity, smart-contract state, access patterns, or embodiment-specific visual evidence from an adversary or inference system (Urzhumov et al., 2012, Avanzini et al., 2019, Inci et al., 2018, Piseno et al., 22 Jun 2026).

1. Transformation-wave cloaks and non-singular generalizations

A central lineage of cloak research comes from transformation optics and transformation acoustics, where the form-invariance of the governing wave equation is used to map a simple “virtual” space into a “physical” space containing a hidden region. In “Isotropic-medium three-dimensional cloaks for acoustic and electromagnetic waves” (Urzhumov et al., 2012), the cloak is a spherical shell composed of only isotropic media, operates in the transmission mode, and requires no mirror or ground plane. Its design is directional rather than omnidirectional: it may reduce visibility of an arbitrary object only for a very limited range of observation angles, but in the short-wavelength limit it restores not only the trajectories of incident rays, but also their phase, which is a necessary ingredient to complete invisibility. The same scalar-index construction is presented for both scalar-wave acoustics and transverse vector-wave electromagnetics.

A second strand regularizes the singular inner boundary of ideal transformation cloaks. “Non-singular arbitrary cloaks dressing three-dimensional anisotropic obstacles” (Dupont et al., 2010) starts from a small region of complex shape instead of a point and derives a transformation matrix for a surface of revolution and its associated non-singular cloak. The resulting cloak avoids vanishing eigenvalues inside the cloak, although the eigenvalues suffer a discontinuity on the inner surface, and all three eigenvalues are independent upon the radius in the concealed object. This replaces the singular “blow-up-a-point” construction by a bounded, though still anisotropic, parameter set.

The same regularization strategy also enables what the paper “Non-singular cloaks allow mimesis” calls mimesis (Diatta et al., 2010). There, non-singular cloaks make objects scatter waves like objects with smaller size and very different shapes. In the Schrödinger setting, a source located inside the cloak scatters waves as if it were located some distance away from a small object, the invisibility region actually hosts almost-trapped eigenstates, and mimetism breaks down for the quantified energies associated with confined modes. The paper further states that non-isomorphic transformations lead to quantum super-scatterers, in which a small object surrounded by a cloak scatters matter waves like a larger nano-object of different shape.

2. Optical implementations and geometric design strategies

Practical optical cloaks have followed several distinct design routes. One route is a nanostructured carpet cloak. “Silicon nanostructure cloak operating at optical frequencies” (0904.3508) demonstrates a cloak operating in the near infrared at a wavelength of 1550 nm. It is a ground-plane cloak that conceals a deformation on a flat reflecting surface, has an area of $225\ \mu\mathrm{m}^2$, hides a region of $1.6\ \mu\mathrm{m}^2$, and is composed of nanometre size silicon structures with spatially varying densities across the cloak. The density variation is defined using transformation optics to define the effective index distribution of the cloak.

A simpler route keeps the medium homogeneous and realizes the required anisotropy through subwavelength layering. “Homogeneous optical cloak constructed with uniform layered structures” (Zhang et al., 2011) reports the experimental realization of a homogenous invisibility cloak with a uniform silicon grating structure. The design eliminates the need for spatial variation of the material index and allows a very large obstacle/cloak ratio; the details explicitly state a ratio of about 0.23 in area. Because the cloak uses a uniform layered medium rather than resonant metamaterials, the reported invisibility behavior is broadband at near-infrared frequencies.

Analytical conformal designs clarify why some carpet cloaks truncate well and others do not. “Conformal carpet and grating cloaks” (Schmied et al., 2010) introduces conformal versions of the quasi-conformal carpet cloak and provides exact refractive-index profiles in closed mathematical form for the usual carpet cloak as well as for other shapes. Its key asymptotic result is that the performance of finite-size cloaks becomes much better for metal shapes with zero average value, such as gratings. This makes grating cloaks a distinguished subclass of carpet cloaks rather than a mere geometric variant.

The visible regime has also been reached with natural birefringent crystals. “Macroscopic Invisibility Cloak for Visible Light” (Zhang et al., 2010) uses two pieces of calcite to conceal a macroscopic object with a maximum height of 2 mm, larger than 3500 free-space-wavelength, inside a transparent liquid environment. Its working bandwidth encompassing red, green and blue light is demonstrated, and the construction uses low-cost materials and simple manufacturing techniques rather than metamaterial nanofabrication.

A further implementation replaces bulk transformation media by guided-wave routing. “Invisibility Cloak Printed on a Photonic Chip” (Feng et al., 2016) realizes a three-dimensional arrayed-waveguide cloak in borosilicate by femtosecond laser direct writing and prototypes up to 5000 waveguides to conceal millimeter-scale volume. The paper characterizes the device by normalized cross correlation, tomography analysis and continuous three-dimensional viewing angle scan, and the details report $\Gamma \approx 0.986$ for the transmitted “$\Psi$” pattern together with an approximately $8.0^\circ$ lateral and $6.7^\circ$ vertical viewing range. In this setting, “invisibility” means that an inner $\hbar$-shaped structure does not measurably affect the transmitted image at the output facet.

3. Diffusive, transport, and static-field cloaks

Cloaking is not restricted to propagating electromagnetic or acoustic waves. In steady heat conduction, “Creation of Tunable Homogeneous Thermal Cloak with Constant Conductivity” (Han et al., 2013) shows that a thermal cloak can be achieved with homogeneous and finite conductivity only employing naturally available materials. The paper emphasizes controlled localization of thermal distribution inside the coating layer and argues that an incomplete cloak can function perfectly. Here the operational criterion is that the temperature and heat flux outside the cloak are indistinguishable from those in the absence of the cloaked region.

“Chemical Cloaking” (Avanzini et al., 2019) extends the idea to reaction-diffusion systems. The cloak is an active chemical shell that, by controlling the concentration of some species in its immediate surrounding, redirects the gradient as if the object was not there. The paper also shows that a substantial fraction of the energy required to cloak can be extracted from the chemical gradient itself, so the cloak is an explicitly non-equilibrium device driven by local chemical work and diffusive free energy.

The transport of overdamped particles can also be cloaked. “Topologically cloaked magnetic colloidal transport” (Rossi et al., 23 Feb 2025) considers paramagnetic colloids driven above a deformed periodic magnetic pattern. There exist topological loops where the particles avoid to trespass the cloaked regions by robustly traveling around the cloak, and afterwards the ensemble of particles continues with a motion identical to the motion as if the distorted region were nonexistent and the ensemble would have trespassed the undeformed region. The paper states a geometric scalability condition: a cloak is scalable to arbitrary size if the biholomorphic map from the undistorted periodic lattice to the region outside the cloak locally rotates by less than an angle of forty five degrees. The work explicitly presents itself as a generalization of cloaking from waves toward particles.

A static-field analogue appears in accelerator physics. “A Magnetic Field Cloak For Charged Particle Beams” (Capobianco-Hogan et al., 2017) describes a passive bilayer cloak combining 45 layers of YBCO high-temperature superconductor with a ferromagnetic shell made from epoxy and stainless steel powder. The reported performance is more than 99% shielding of a transverse magnetic field of up to 0.45 T and 95% shielding at 0.5 T at liquid nitrogen temperature, while the ferromagnetic shell reduces field distortions caused by the superconductor alone by 90% at 0.45 T. Here cloaking means simultaneously creating a nearly field-free beam channel and leaving the surrounding magnetic field nearly undisturbed.

4. Performance criteria, approximations, and controversy

The literature distinguishes sharply between different notions of “hiding.” In the eikonal formulation of “Isotropic-medium three-dimensional cloaks for acoustic and electromagnetic waves” (Urzhumov et al., 2012), a cloak must achieve both ray restoration and phase restoration. The same paper argues that regions with $n<1$ are essential because a detoured ray otherwise accrues extra phase; isotropy removes birefringence but forces a trade-off toward directionality and the short-wavelength limit. This makes the device approximate and unidirectional rather than omnidirectional and exact.

A conceptually different route is active cancellation. “Active Cloaking” (Selvanayagam et al., 2012) introduces an array of sources that cancels electromagnetic scattering by superimposing magnetic and electric surface current densities at the boundary of an object. These currents can be discretized into electric and magnetic dipoles implemented by straight and loop wire antennas, and the paper verifies the approach numerically for both dielectric and metallic cylinders. The criterion is again external-field restoration, but the mechanism is source synthesis rather than passive transformation media.

The evaluation metric itself is contested. In “Comment on ‘A self-assembled three-dimensional cloak in the visible’” (Miller et al., 2013), Miller and coauthors argue that reducing scattering alone is not sufficient if absorption increases so much that total extinction rises. Their calculation shows that for the criticized nanoparticle-coated silica sphere, total extinction increases at all wavelengths between 250 and 500 nm; at $\lambda \approx 335$ nm the apparent size increases by almost a factor of three, and over 300–400 nm the smallest enhancement factor of $C_{\text{ext}}$ is about 1.65. The paper states the point bluntly: an object creating a large shadow is generally not considered to be cloaked.

Analytical asymptotics reinforce why some approximations fail gracefully and others do not. “Conformal carpet and grating cloaks” (Schmied et al., 2010) shows that finite-size performance depends on how the refractive index approaches its background value. Metal shapes with zero average value, such as gratings, lead to much better finite-size cloaks than one-sided bumps, because the relevant asymptotic decay is far more favorable under truncation.

5. Cloak in security, privacy, and confidential computation

Outside wave physics, “cloaking” has become a technical term for hiding semantically informative traces from an adversary. “DeepCloak: Adversarial Crafting As a Defensive Measure to Cloak Processes” (Inci et al., 2018) uses adversarial learning as a defensive tool against side-channel classification based on Hardware Performance Counters. The paper first trains highly accurate models, reaching up to 99.8% validation accuracy for process classification, and then shows that minimal perturbations to leakage traces can cloak a process from a malicious classifier. Its abstract states that even in the presence of adversarial re-training and defensive distillation, all 10 of the tested adversarial learning methods still manage to successfully craft adversarial perturbations and the proposed cloaking methodology succeeds.

In blockchain systems, “CLOAK: A Framework For Development of Confidential Blockchain Smart Contracts” (Ren et al., 2021) addresses confidentiality of smart contracts through trusted execution environments. The framework introduces a domain-specific annotation language for declaring privacy specifications and automatically generates confidential smart contracts to be deployed with TEE on blockchain. Its key target is the multi-party transaction problem, and the evaluation reports that developers managed to deploy business services on blockchain in a concise manner by only developing CLOAK smart contracts whose size is less than 30% of the deployed ones.

In oblivious storage, “Cloak: Heuristic ORAM Optimization Through Fixed Temporal Distribution” (Arpaci et al., 26 May 2026) uses a fixed, recentness-biased temporal distribution of server accesses. The paper emphasizes that this heuristic affects only performance, not security, and reports overheads as low as $1.6\ \mu\mathrm{m}^2$0 over a non-oblivious and unencrypted baseline. On Netflix click-stream and Ethereum transaction traces it reports 165,000 and 157,000 operations per second, respectively, on a single machine. In this setting, cloaking means that server-side access patterns remain completely independent of the clients’ real queries even though the schedule is engineered to exploit temporal locality.

6. Embodiment cloaking in robot learning

A recent machine-learning use of the term shifts from hiding an object in the world to hiding the robot from its own policy. “Cloak: Zero-Shot Cross-Embodiment Manipulation by Masking the End-Effector from the VLA” (Piseno et al., 22 Jun 2026) presents a training recipe for Vision-Language-Action models in which the end-effector is cloaked from its own wrist camera. The end-effector occupies a large and consistent region of the wrist view, and masking it allows for embodiment-agnostic visual reasoning. The mask is rendered in simulation from the robot’s known geometry, accurately and in real time, with no segmentation or generative models.

The resulting Cloak-VLA is trained with Cloak on a single parallel-jaw gripper dataset, and no data of new embodiments is ever collected. The paper reports zero-shot transfer to various unseen embodiments, including another gripper, another arm, and a five-fingered hand, while preserving the source embodiment’s performance (Piseno et al., 22 Jun 2026). The mechanism is not optical concealment but visual invariance: by decoupling the wrist view from its own embodiment, Cloak allows data to outlive the hardware it was collected on.

This suggests a broader semantic shift in the modern literature. In older physical uses, a cloak modifies fields so that a hidden region does not perturb observation. In contemporary robotic and computational uses, a cloak modifies observations, traces, or representations so that an observer—or a model acting as an observer—cannot reliably infer the identity of the underlying process, state, access pattern, or embodiment (Inci et al., 2018, Ren et al., 2021, Arpaci et al., 26 May 2026, Piseno et al., 22 Jun 2026).