---
title: Soft Continuum Robot-Inspired Antenna Array
url: https://www.emergentmind.com/topics/soft-continuum-robot-inspired-antenna-array
type: topic
---

# Soft Continuum Robot-Inspired Antenna Array

Searching arXiv for the specified works and closely related papers to ground the article.
arXiv search query: 2507.23719 OR 2507.06589 OR 2606.11771 OR 2505.09870 OR 2308.07458
Soft continuum robot-inspired antenna arrays are antenna architectures in which the array geometry is reconfigured through deformation of a compliant, tentacle-like, or segment-wise soft structure rather than by moving each radiating element independently. In the communications literature, this idea is formalized as a base-station array with multiple flexible tentacles whose shapes are controlled by a small set of geometric parameters, so that the positions of many antenna elements are changed coherently at the structure level [2507.06589]. A later segment-wise formulation extends this principle to soft robotic arms with independently controllable sections that support bending, elongation–retraction, and sweeping motions, together with alternative deployment schemes for end-mounted and intra-segment antennas [2606.11771]. Related work on dynamically stabilized flexible phased arrays, deployable co-cured apertures, and robophysical tactile antennae provides material systems, fabrication strategies, and morphology-aware sensing paradigms that are closely aligned with the broader concept of a deformable continuum-inspired antenna system [2505.09870], [2308.07458], [2507.23719].

## 1. Conceptual definition and scope

In its most specific communications formulation, a soft continuum robot-inspired antenna array is a **multi-tentacle antenna structure** equipped with a total of \(MN\) radiating elements, where each tentacle is a continuum structure of fixed arc length \(L_{\max}\) and its 3D shape is controlled by an amplitude \(A_m\) and a spatial frequency \(v_m\) [2507.06589]. The defining distinction is that **reconfigurability is achieved at the structure level**. This contrasts with **conventional rigid arrays**, where geometry is fixed and only digital beamforming or precoding is used, and with **per-element reconfigurable arrays**, where each element is moved or switched individually and therefore incurs many control variables and actuators [2507.06589].

The segment-wise extension makes the continuum interpretation more explicit. In that formulation, the base station consists of \(M\) soft robotic **tentacles**, each partitioned into \(S\) serial **segments**, and each segment can independently perform **bending**, **elongation–retraction**, and **sweeping** motions [2606.11771]. The paper describes this as a **segment-wise soft robotic antenna (SRA) system**, in which the antennas are mounted on the **surface** of the soft body. Two deployment schemes are introduced: the **segmented end-antenna configuration (SEAC)**, where fixed antennas are mounted at the segment ends and reconfigured via segment motions, and the **hybrid end-and-intermediate antenna configuration (HEIAC)**, where reconfigurable antennas are further integrated as intra-segment antennas [2606.11771].

A recurrent misconception is that such arrays are simply another instance of per-element reconfigurable antennas. The primary literature states the opposite: the soft continuum approach reduces the number of control parameters because a tentacle shape is controlled by a few structural variables, while many mounted elements move coherently with that deformation [2507.06589]. A second misconception is that the concept requires a mathematically continuous backbone. The robophysical and segment-wise literature instead treats discretized compliant structures as valid continuum approximations when the deformation remains smooth and connectivity constraints are enforced [2507.23719], [2606.11771].

## 2. Geometric and kinematic models

The initial continuum model places \(M\) tentacles in equally spaced azimuthal directions,
\[
\theta_m = \frac{2\pi m}{M},
\]
and parameterizes the \(m\)-th tentacle by arc length \(\ell \in [0,L_{\max}]\). Its 3D position is
\[
\mathbf{r}_m(\ell,t)=
\begin{bmatrix}
u_m(\ell,t)\cos\theta_m\\
u_m(\ell,t)\sin\theta_m\\
A_m\sin(\omega t+v_m\ell)
\end{bmatrix},
\]
where \(A_m\) is the deformation amplitude and \(v_m\) is the spatial frequency [2507.06589]. For the static optimization problem, the explicit time dependence is dropped, yielding
\[
z_m(\ell)=A_m\sin(v_m\ell).
\]
The constant-arc-length requirement imposes
\[
\left\Vert \frac{\partial \mathbf r_m}{\partial \ell}\right\Vert
=
\sqrt{\left(\frac{\partial u_m}{\partial \ell}\right)^2+
\left(\frac{\partial z_m}{\partial \ell}\right)^2}=1,
\]
so that
\[
\frac{\partial u_m}{\partial \ell}
=
\sqrt{1-\left(A_m v_m \cos(v_m\ell)\right)^2},
\qquad
u_m(\ell)
=
\int_0^\ell
\sqrt{1-\left[A_m v_m \cos(v_m\bar\ell)\right]^2}\,d\bar\ell.
\]
The condition \(|A_m v_m|\le 1\) is imposed to ensure that the geometry remains physically meaningful [2507.06589].

Antenna elements are then sampled along the tentacle at
\[
\ell_n=\frac{n}{N}L_{\max},
\]
with coordinates
\[
\mathbf r_{m,n}=
\begin{bmatrix}
u_m(\ell_n)\cos\theta_m\\
u_m(\ell_n)\sin\theta_m\\
A_m\sin(v_m\ell_n)
\end{bmatrix}.
\]
This model already captures a key continuum-robotic effect: when \(A_m\neq 0\), the tentacle retracts in the \(xy\)-plane while bending in \(z\), so structural deformation changes the full 3D aperture geometry rather than only the elevation profile [2507.06589].

The segment-wise formulation replaces each tentacle by a serial chain of independently actuated soft segments. For segment \(s\) of tentacle \(m\), the pointwise model is
\[
\mathbf r_{m,s}(\ell)=
\begin{bmatrix}
\ell\cos\theta_m\\
\ell\sin\theta_m\\
A_{m,s}\sin(v_{m,s}\ell)
\end{bmatrix},
\]
and the corresponding arc-length mapping is
\[
L(\ell)=\int_0^\ell
\sqrt{1+\left(A_{m,s}v_{m,s}\cos(v_{m,s}\bar\ell)\right)^2}\,d\bar\ell.
\]
The endpoint arc lengths \(L_{m,s}\) encode elongation–retraction, and inter-segment smoothness is enforced through \(\mathcal C^0\) and \(\mathcal C^1\) continuity constraints at the segment boundaries [2606.11771]. The final antenna coordinates are
\[
\mathbf r_{m,s,n}=
\begin{bmatrix}
\ell(L_{m,s,n})\cos\theta_m\\
\ell(L_{m,s,n})\sin\theta_m\\
A_{m,s}\sin\!\big(v_{m,s}\ell(L_{m,s,n})\big)
\end{bmatrix}.
\]
This segment-wise construction is described as a **discrete approximation** of a continuum robot, and the continuity constraints function as a communication-oriented analog of soft-robot kinematic regularity [2606.11771].

## 3. Electromagnetic and communication models

The communications model in the 2025 formulation is a multi-user multiple-input single-output downlink system. Each element is assigned a directional cosine pattern,
\[
Q_E(\vartheta,\varphi)=
\begin{cases}
Q\cos^\kappa\vartheta, & \vartheta\in[0,\pi/2],\ \varphi\in[0,2\pi],\\
0, & \text{otherwise},
\end{cases}
\]
with \(Q=2(\kappa+1)\), and the array steering response is
\[
\left[\mathbf a_m(\vartheta,\varphi)\right]_n
=
\exp\!\Big(
-\frac{j2\pi}{\lambda}
\big(
x_{m,n}\sin\vartheta\cos\varphi+
y_{m,n}\sin\vartheta\sin\varphi+
z_{m,n}\cos\vartheta
\big)
\Big).
\]
The channel is represented by a geometric Saleh–Valenzuela model with \(N_c\) clusters and \(N_p\) paths per cluster, so the deformation parameters enter the channel through the element coordinates and therefore through the steering vectors [2507.06589].

The transmit signal uses a zero-forcing precoder,
\[
\mathbf F=\mathbf H^H(\mathbf H\mathbf H^H)^{-1},
\qquad
\alpha=\sqrt{\frac{P_{\max}}{\operatorname{Tr}(\mathbf F\mathbf F^H)}},
\qquad
\mathbf W=\alpha\mathbf F,
\]
and the user SINR is
\[
\gamma_k(\mathbf z)=
\frac{|\mathbf h_k^H(\mathbf z)\mathbf w_k|^2}
{\sum_{i\neq k}|\mathbf h_k^H(\mathbf z)\mathbf w_i|^2+\sigma^2}.
\]
The optimization objective is the sum rate
\[
R(\mathbf z)=\sum_{k=1}^K \log_2(1+\gamma_k(\mathbf z)),
\]
subject to \(0\le A_m\le A_{\max}\), \(0\le v_m\le v_{\max}\), and \(|A_m v_m|\le 1\) [2507.06589].

Because this problem is non-convex, the paper introduces a **successive convex approximation** method. The reduced variable vector is
\[
\bar{\mathbf z}=[A_1,\dots,A_M,v_1,\dots,v_M]^T,
\]
and the sum rate is linearized around the current iterate by a first-order Taylor approximation. The nonlinear constraint \(|A_m v_m|\le 1\) is also linearized as
\[
\left|
A_m v_m^{(i)} + A_m^{(i)} v_m - A_m^{(i)}v_m^{(i)}
\right|
\le 1.
\]
This converts each step into a convex subproblem while preserving the structure-level parameterization [2507.06589].

The 2026 segment-wise formulation changes both the array model and the communication setting. It considers an uplink MU-SIMO system with a spatially correlated Rayleigh channel,
\[
\mathbf h_k(\mathbf r)\sim \mathcal{CN}(\mathbf 0,\sigma_k^2\mathbf C(\mathbf r)),
\]
MMSE combining,
\[
\mathbf W=
\left(\mathbf H^H(\mathbf r)\mathbf H(\mathbf r)+\frac{1}{\gamma}\mathbf I\right)^{-1}\mathbf H^H(\mathbf r),
\]
and user SINR
\[
\gamma_k(\mathbf r)=
\frac{|\mathbf w_k\mathbf h_k(\mathbf r)|^2}
{\sum_{g\in\mathcal K\setminus\{k\}}|\mathbf w_k\mathbf h_g(\mathbf r)|^2+\sigma^2\|\mathbf w_k\|^2}.
\]
The spatial correlation coefficients are expressed by 3D Clarke-type sinc kernels based on Euclidean antenna separations, making the segment geometry directly responsible for correlation control [2606.11771].

For SEAC, the sum-rate maximization with continuity constraints is solved by a **penalty dual decomposition-projected gradient ascent (PDD-PGA)** algorithm. For HEIAC, the paper jointly optimizes segment deformation, intra-segment antenna positions, and antenna activation using a **block coordinate descent (BCD)-PDD-PGA algorithm with greedy backward antenna selection** [2606.11771]. The architecture therefore combines soft-body kinematics, correlation-aware placement, and mixed continuous-discrete optimization.

## 4. Physical realizations, materials, and morphing hardware

Although the communications papers are principally modeling and optimization studies, several related hardware papers supply practical substrates for deformable array realization. One example is a tiled, additively printed flexible \(4\times 4\) array at \(2.1\) GHz composed of four \(2\times 2\) tiles, each with printed circular patch antennas, a CMOS beamforming integrated circuit, and a local **Dynamic Beam-Stabilized (DBS)** processor [2505.09870]. The DBS processor performs beam adaptation through on-chip real-time control of gain, phase, and delay for each element, and uses a perturb-and-observe style loop based on the beamformed output. In deformation experiments at radius \(R=38\) cm, the dynamic beam pointing error was reduced from about \(7^\circ\) without correction to less than \(1.5^\circ\) after convergence; the abstract reports \(1.25^\circ\) [2505.09870].

That same system uses **Copper Molecular Decomposition (CuMOD)** ink, reported to have **< 0.1% variation per degree C with temperature and strain**, together with a tile architecture described as low power, low-area, and easily scalable [2505.09870]. A plausible implication is that a soft continuum array could distribute analogous local beam-stabilization electronics along a deforming backbone, allowing each region of the aperture to compensate its own geometric perturbation without requiring a global deformation codebook.

A second hardware lineage is the \(10\) GHz **popup array**, a light and flexible array composed of dipole antennas co-cured to a glass-fiber composite [2308.07458]. Each element is a mechanically self-deploying dipole radiator built as a co-cured electromechanical laminate. The structure can fold completely flat, coil, and pop back up upon deployment, and the demonstrator passed vibration, thermal, high-frequency thermal cycling, thermal shock, and prolonged stowage tests [2308.07458]. The measured ensemble center frequency is \(10.4032\) GHz, the mean bandwidth is \(1.8285\) GHz, the broadside gain is \(5.3\) dBi, and the half-power beam width is approximately \(110^\circ\) in both principal planes [2308.07458].

These two hardware directions are not identical to a soft continuum robot-inspired array, but they show two complementary routes to realization: one based on flexible electronics plus dynamic RF correction, and one based on compliant co-cured backbones plus deployable morphology. The former emphasizes real-time beam stabilization under deformation; the latter emphasizes foldability, shape repeatability, and environmental robustness [2505.09870], [2308.07458].

## 5. Robophysical tactile antennae and embodied morphology

The term *antenna* in soft-robotic literature also appears in a tactile, rather than RF, sense. The paper **"Design of a bioinspired robophysical antenna for insect-scale tactile perception and navigation"** introduces CITRAS, a **Cockroach Inspired Tactile Robotic Antenna Sensor**, explicitly described as a robophysical analogue of the American cockroach antenna and best viewed as a **discretized continuum arm with embedded joint sensing** at insect scale [2507.23719]. CITRAS has nine rigid segments connected by eight compliant flexural hinges, a linearly decreasing hinge width from \(8.0\) mm at the base to \(3.62\) mm at the tip, and a segmented compliant structure that passively bends in response to environmental stimuli [2507.23719].

The sensing principle is based on embedded sliding parallel-plate capacitors at the hinges, with theoretical angle sensitivity
\[
S_\theta^{\text{th}}\approx 7.14\ \mathrm{fF/deg},
\]
an experimentally calibrated sensitivity of about \(26.5\ \mathrm{fF/deg}\) on hinge H1, and per-hinge 3rd-order polynomial calibration with \(R^2>0.99\) in the operational range [2507.23719]. The prototype is compact \((73.7\times 15.6\times 2.11\ \mathrm{mm})\), lightweight \((491\ \mathrm{mg})\), and low-power \((32\ \mathrm{mW})\). Reported errors include \(0.056\pm 0.079^\circ\) mean quasi-static angle reconstruction error, a maximum quasi-static error up to \(0.795^\circ\), \(0.10\pm 0.27^\circ\) mean dynamic bending error below saturation, base-to-tip distance error up to \(7.75\%\), and gap-width errors of \(6.73\%\), \(5.11\%\), and \(4.66\%\) for predicted widths \(22.68\), \(50.45\), and \(78.23\) mm [2507.23719].

This tactile branch is not an RF array, but it is conceptually relevant because it demonstrates the same structural ideas: stiffness gradients, passive conformation, distributed sensing, and coarse spatial discretization of a continuum backbone. The paper states that, from a continuum robotics perspective, the structure approximates a continuous curvature backbone, and that the angle readings can be used as a coarse spatial discretization of continuous curvature [2507.23719]. A plausible implication is that soft continuum robot-inspired RF arrays and robophysical tactile antennae share a common design language centered on morphology as a computational and sensing resource.

## 6. Performance, trade-offs, and open problems

The principal communications benefit claimed for structure-level deformation is sum-rate improvement. In the 2025 multi-user MISO formulation, the proposed deformable array significantly outperforms fixed geometry and per-element reconfigurable arrays in sum rate, with reported gains of up to **73%** over fixed geometry and **26%** over the 2D reconfigurable concentric circular antenna array baseline [2507.06589]. At \(K=2\) and SNR \(=20\) dB, the soft robot antenna achieves **61.9% higher sum rate** than fixed CCAA and **18.1% higher sum rate** than 2D reconfigurable CCAA, while the hypothetical 3D reconfigurable CCAA still exceeds it by **11.1%** [2507.06589]. This comparison clarifies a central trade-off: the continuum array reduces mechanical and control complexity, but it does not dominate a fully free 3D per-element benchmark in every regime.

The segment-wise formulation reports even stronger results against conventional 3D reconfigurable arrays. SEAC and HEIAC achieve **37.9%** and **32.1%** sum-rate gains over conventional 3D reconfigurable arrays, respectively, and SEAC provides up to a **49.3%** gain in compact array deployments [2606.11771]. The same study also shows that, in HEIAC with \(N=11\) candidate antennas per segment, the greedy backward algorithm can deactivate \(5\) intra-segment antennas when \(\eta\in[0.5,0.9]\) and \(6\) when \(\eta\in[1.0,1.2]\), while at \(\eta=1.2\) a reduction from \(69.2091\) to \(67.4763\) bps/Hz corresponds to only about a \(2.5\%\) sum-rate decrease [2606.11771]. These results underscore the importance of activation control when dense local placement creates strong intra-segment correlation.

At the same time, the literature identifies several unresolved issues. The 2025 DBS work experimentally validates uniform bending and dynamic beam recovery, but twisting and non-uniform 3D bending are not experimentally validated, and convergence speed under actual continuous deformation is not explicitly quantified [2505.09870]. The tactile CITRAS platform is planar-only in its current design, has a dynamic range limited to about \(\pm 10^\circ\) per hinge before saturation, and exhibits low damping with \(\zeta \approx 0.035 \pm 0.012\), substantially below the cited biological value \(\zeta\sim 0.3\) [2507.23719]. The segment-wise SRA papers are communication-oriented and therefore rely on reduced-order sinusoidal deformation models rather than full continuum mechanics; this suggests that future work on calibration, actuation fidelity, and mutual-coupling-aware co-design remains necessary [2606.11771].

Taken together, these papers define a coherent research area. A soft continuum robot-inspired antenna array is not merely a flexible substrate carrying antennas; it is a mechanically reconfigurable electromagnetic system in which compliant-body kinematics, antenna placement, channel shaping, and control are jointly exploited. Across the available literature, the concept spans structure-level geometry optimization for wireless sum rate, local deformation-aware beam stabilization, deployable compliant apertures, and robophysical tactile antennae that treat deformation itself as a sensing modality [2507.06589], [2505.09870], [2308.07458], [2507.23719], [2606.11771].

Source: https://www.emergentmind.com/topics/soft-continuum-robot-inspired-antenna-array