Multi-Magnet Embedded Soft Continuum Robots

• MeSCRs are soft continuum robots that embed permanent magnets along their structure to decouple bending and torsion for flexible actuation.
• Their design leverages radial and distributed magnet layouts to optimize task-space degrees of freedom and enable mode switching for applications like targeted drug delivery.
• Recent studies integrate reduced-order modeling and energy-based optimization to predict magneto-elastic behavior and drive precise, application-specific design.

Searching arXiv for the cited MeSCR papers and closely related continuum/magnetic soft robotics work. Multi-Magnet Embedded Soft Continuum Robots (MeSCRs) are slender soft continua—typically catheter- or guidewire-like bodies—in which multiple permanent magnets or hard-magnetic regions are embedded along the backbone or radially within the wall and actuated by externally generated magnetic fields. Their defining feature is multi-point magnetic torque generation: instead of relying on a single distal magnet or a uniformly magnetized shaft, MeSCRs distribute magnetic actuation over multiple embedded dipoles to induce bending, torsion, or both. Recent work has formalized MeSCRs as magneto-elastic continuum systems with analyzable equilibrium structure, controllability limits, and layout-dependent kinematic performance, while also demonstrating application-specific embodiments such as a dual-mode magnetic continuum robot for targeted drug delivery (Zhang et al., 2 Oct 2025) and a unified modeling framework in which the maximum controllable task-space degrees of freedom is bounded by twice the number of embedded magnets (Wu et al., 15 Jul 2025).

1. Conceptual scope and actuation regimes

MeSCRs are modeled as soft, slender, typically inextensible continua with multiple embedded permanent micromagnets or hard-magnetic inclusions distributed along the backbone. In the unified framework, the backbone is an isotropic soft matrix with Young’s modulus $E$ and shear modulus $G = E/[2(1+\rho)]$, while the embedded magnetic elements have dipole magnitudes $M_k$ and occupy a subset of discrete backbone locations. External magnetic fields generate torques on each embedded dipole, and these torques collectively induce bending and twist of the backbone under quasi-static equilibrium assumptions (Wu et al., 15 Jul 2025).

A central distinction within the broader magnetic continuum robot literature is between axially magnetized magnetic continuum robots and MeSCRs with multiple, non-collinear embedded dipoles. Axially magnetized designs have largely been associated with bending-dominant behavior. By contrast, radial or otherwise spatially distributed multi-magnet layouts are used specifically to enlarge the deformation basis and to decouple deformation modes. In the dual-mode catheter embodiment, the motivation is explicit: radial embedding is introduced to unlock both bending and torsion with a single externally steered permanent magnet and to couple torsional deformation to a compact drug-release mechanism (Zhang et al., 2 Oct 2025).

Controllability is not determined solely by the number of magnets. The unified actuation-rank result gives

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$

so the maximum controllable task-space degrees of freedom is $2N_m$. This yields the practical regimes $N_m=1$ for at most $2$ DoF, $N_m=2$ for at most $4$ DoF, $N_m=3$ for at most full $G = E/[2(1+\rho)]$0-DoF actuation, and $G = E/[2(1+\rho)]$1 for redundancy. However, if the field is spatially uniform and shared, only three independent field channels exist, so controllable DoF is limited to $G = E/[2(1+\rho)]$2 regardless of $G = E/[2(1+\rho)]$3. A common misconception is therefore that adding magnets alone guarantees high-rank control authority; the field manifold is equally decisive (Wu et al., 15 Jul 2025).

2. Embedding architectures and mechanical layouts

One MeSCR architecture embeds multiple permanent magnets radially at the distal tip of a soft catheter. In the demonstrated prototype, three internal permanent magnets are arranged around the circumference, with the intervening spaces filled by solid material to lock their orientations and spacing. A nitinol wire runs internally, and an “Internal Block” is rigidly attached to both the nitinol wire and the fixed end of the catheter using solid filler so that the segment between the release groove and the fixed end does not undergo axial torsion and remains mechanically decoupled from the catheter wall. The same platform incorporates a dual-layer blockage mechanism: outer grooves are machined on the catheter wall near the distal tip, and inner plates form a compact Internal Block that occludes the groove until twist-induced relative motion opens the outlet. For the phantom study, the Internal Block was fabricated by micro-nano 3D printing with geometry specified as a cylinder of radius $G = E/[2(1+\rho)]$4 mm and height $G = E/[2(1+\rho)]$5 mm plus a cylindrical sector of the same radius, height $G = E/[2(1+\rho)]$6 mm, and central angle $G = E/[2(1+\rho)]$7. An air segment is intentionally preserved between the drug reservoir and the blockage to prevent unintended leakage until torsion is applied (Zhang et al., 2 Oct 2025).

Another MeSCR architecture is the magnetic ball chain robot, which realizes a “multi-lumped magnet” design: a chain of permanently magnetized spheres is enclosed in a cylindrical polymer skin. In the comparative study, this architecture used chains of $G = E/[2(1+\rho)]$8 mm diameter N52 spheres in simulation, while experiments used ten N42 spheres of diameter $G = E/[2(1+\rho)]$9 mm, mass $M_k$0 g, and remanence $M_k$1 T. The spheres form spherical joints through direct contact, and a thin Ecoflex 00-30 sleeve provides sealing and small distributed elastic bending resistance. For the $M_k$2 mm outer-diameter design, the sleeve thickness is $M_k$3 mm around $M_k$4 mm spheres, with silicone modulus $M_k$5 kPa. This separates magnetic content from backbone bending stiffness: magnetic material is concentrated in the spheres, while bending occurs primarily at inter-sphere contacts rather than through a stiff magnetized backbone (Pittiglio et al., 2023).

These embodiments delimit two major design philosophies. Radial magnet embedding localizes multiple dipoles at the distal tip to generate mode switching between bending and torsion and to actuate distal mechanisms. Ball-chain designs distribute many discrete dipoles along the backbone to maximize steerability and local curvature. A plausible implication is that “MeSCR” is best understood as a family of architectures defined by multi-point embedded magnetic actuation rather than by a single canonical geometry.

3. Magneto-mechanical principles and deformation decoupling

At the local level, each embedded magnet is modeled as a magnetic dipole with moment vector $M_k$6, volume $M_k$7, and magnetization $M_k$8, so that

$M_k$9

For an external permanent magnet with magnetic moment $\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$0 located at position $\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$1 relative to an embedded magnet, the dipole field is

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$2

with $\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$3. The magnetic torque and force on embedded magnet $\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$4 are

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$5

In the MeSCR formulations considered here, torque is usually dominant, while force is typically much smaller; however, force can become non-negligible at very short separations due to field gradients (Zhang et al., 2 Oct 2025).

The dual-mode catheter exploits field orientation to decouple deformation modes. In bending mode, the external field is aligned with the distal axis so that the magnetic torques are misaligned with the catheter’s central axis, producing a net bending moment within a defined bending plane while suppressing torsional torque on the cross-section. In torsion mode, the external field is oriented coplanar with the catheter cross-section, so that the magnetic torque on each embedded magnet aligns with the catheter’s central axis and generates net twist. The underlying continuum mechanics are expressed through the Euler–Bernoulli relation

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$6

for bending and the Saint-Venant/Timoshenko torsion relation

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$7

for torsion. In the three-magnet bending model, the paper gives the angle-dependent torque magnitude

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$8

and the tip curvature superposition

$\operatorname{rank}(J_b) \le \min\{6,\,2N_m\},$9

which makes explicit the $2N_m$0 scaling of magnetic actuation and the inverse dependence on flexural rigidity $2N_m$1 (Zhang et al., 2 Oct 2025).

The same catheter uses torsion to actuate drug release. The external tube is divided into a distal segment $2N_m$2 between tip and release groove and a proximal segment $2N_m$3 between release groove and fixed end. With distal magnetic torque $2N_m$4, proximal twist torque $2N_m$5, and distributed resisting torque $2N_m$6 in the $2N_m$7 segment, the internal torsion moment is

$2N_m$8

and the twist angle is

$2N_m$9

The relative twist that disengages the groove from the Internal Block is

$N_m=1$0

with resulting sliding distance

$N_m=1$1

This directly yields the design rule that increasing $N_m=1$2 amplifies sliding for fixed $N_m=1$3 and friction $N_m=1$4 (Zhang et al., 2 Oct 2025).

In ball-chain MeSCRs, the same magneto-mechanical logic is written as total potential energy minimization. The total energy combines external field interaction, dipole–dipole interaction between spheres, elastic sleeve bending, and optionally gravity:

$N_m=1$5

The equilibrium configuration minimizes $N_m=1$6 subject to contact constraints between adjacent spheres. This discrete energy view is particularly natural for architectures whose curvature localizes at joint-like contacts rather than being smoothly distributed (Pittiglio et al., 2023).

4. Reduced-order models, rod theories, and structural optimization

A general analytical treatment of MeSCRs is provided by the extended pseudo-rigid-body model (EPRB), in which the continuous backbone is approximated by $N_m=1$7 rigid rods linked by $N_m=1$8 spherical joints, with compliant elements concentrated at the joints. The generalized coordinates are $N_m=1$9, where each joint variable $2$0 parameterizes rotations about local axes. The total potential energy is

$2$1

with quadratic elastic energy

$2$2

and magnetic energy

$2$3

Quasi-static equilibrium satisfies

$2$4

or, in torque-dominant compact form,

$2$5

This formulation yields explicit joint-space compliance, Cartesian compliance, and an actuation Jacobian $2$6 (Wu et al., 15 Jul 2025).

The same framework establishes well-posedness conditions. A sufficient uniqueness condition for the fixed-point equilibrium $2$7 is

$2$8

Boundedness and convergence are obtained through a gradient-flow analysis. For axial magnetization, the theory also proves a material-twist-free result: if all embedded magnets are axially aligned and the field magnitude satisfies a stated bound, the MeSCR is material-twist-free for any actuating field. Under coplanar-field conditions, there exists a planar equilibrium whose backbone centerline lies entirely in a plane. These results formalize why axially magnetized architectures are naturally associated with planar bending rather than torsional behavior (Wu et al., 15 Jul 2025).

Kinematic performance is expressed geometrically. Local actuation manipulability is defined as

$2$9

and the immersion metric

$N_m=2$0

quantifies distortion from actuation space into the configuration manifold. The structural optimality condition

$N_m=2$1

shows that improving local performance requires spectral shaping of $N_m=2$2, specifically counteracting anisotropy and increasing small eigenvalues. Closed-form two-magnet optima were derived in representative planar uniform-field cases: $N_m=2$3 for aligned moments and $N_m=2$4 for opposing moments (Wu et al., 15 Jul 2025).

Rod-theoretic models provide the continuum counterpart to these reduced-order formulations. The review of rod models identifies Cosserat–Reissner rod theory as the most general 3D choice for bending, torsion, shear, and extension; Kirchhoff–Love theory for inextensible, unshearable bend–twist behavior; Euler–Bernoulli for planar bending; and Timoshenko–Ehrenfest for shear-flexible planar problems. Within this framework, magnetic effects enter as external distributed wrenches or discrete point wrenches along arc length. Richter, Venkiteswaran, and Misra are cited for multi-point orientation control of discretely magnetized continuum manipulators using a quasi-static Cosserat model; Tariverdi et al. for dynamic magnetic loading in an extended Kirchhoff rod; and Xiang et al. for Euler–Bernoulli-based prediction of a tetherless magnetic soft robot. The same review surveys inverse statics, inverse dynamics, feedback linearization, energy shaping, sliding-mode control, model predictive control, and learning-based control with rod priors as directly relevant tools for MeSCR control (Alessi et al., 2024).

5. Computational co-design and magnetization programming

MeSCR design has also been cast as a joint structure–magnetization–stimulus optimization problem. In the magneto-elastic material point method (MPM) framework for hard-magnetic soft robots, the reference configuration $N_m=2$5 deforms via $N_m=2$6 with deformation gradient $N_m=2$7 and Jacobian $N_m=2$8. The Helmholtz energy density is

$N_m=2$9

with compressible Neo-Hookean elastic part

$4$0

and magnetic part

$4$1

The first Piola–Kirchhoff stress follows from $4$2, with magnetic contribution

$4$3

Magneto-elastic effects are therefore propagated through stress rather than through an explicitly added body magnetic force (Wang, 28 Mar 2025).

The MPM implementation uses particles carrying mass, volume, position, velocity, affine velocity, deformation gradient, and material properties, and a fixed Eulerian grid storing mass and velocity. Particle-to-grid and grid-to-particle transfers are differentiated end-to-end through the entire solver using auto-differentiation on GPU. The implementation is reported in Taichi on an NVIDIA RTX 6000 Ada with 48 GB memory. Even though velocity overwrites at contacts locally break the chain rule, the paper reports that distributed particle-to-grid and grid-to-particle dependencies preserve most gradient signal and enable robust sensitivities in dynamic contact scenarios (Wang, 28 Mar 2025).

For MeSCRs, multiple embedded magnets are represented as spatially varying residual flux density $4$4 over design regions $4$5. Each region is assigned a normalized remanence magnitude and an orientation vector, while the external field is parameterized as a harmonic stimulus

$4$6

Design variables include relaxed density $4$7, remanence magnitude, remanence orientation, field amplitude, field orientation, and frequency. Single-task and multi-task objectives are then optimized under MPM dynamics using Adam updates with projection to admissible bounds. The reported stopping condition is that the moving average of the objective over 10 iterations changes by less than $4$8 relative to the previous 10 iterations (Wang, 28 Mar 2025).

The framework has been applied to MeSCR-relevant examples. A quasi-static distal continuum beam of size $4$9 mm $N_m=3$0 $N_m=3$1 mm was segmented into $N_m=3$2 design regions, with material parameters $N_m=3$3 kPa, $N_m=3$4, and $N_m=3$5 T. For tip-height maximization, obstacle avoidance, and multi-task shape morphing, optimized remanence patterns varied spatially, often placing stronger remanence near the tip and weaker remanence near the base or flipping remanence near constrained regions. Runtime for the quasi-static cases was reported as $N_m=3$6–$N_m=3$7 iterations at about $N_m=3$8 s per iteration, for totals not exceeding $N_m=3$9 minutes. This suggests that MeSCR design need not be restricted to heuristic magnet placement; it can be posed as a differentiable co-design problem in which geometry, local magnetization, and field schedule are optimized simultaneously (Wang, 28 Mar 2025).

6. Experimental manifestations and comparative performance

The dual-mode magnetic continuum robot provides the most explicit MeSCR demonstration of coupled navigation and functional payload release. The benchtop characterization used an $G = E/[2(1+\rho)]$00 mm-long catheter segment with proximal clamp, three internal permanent magnets embedded radially at the distal tip, two yellow labels for stereo vision, and red indicators to visualize magnetization directions. By varying the distance $G = E/[2(1+\rho)]$01 between the external permanent magnet and the internal magnets, the experiments measured a maximum bending angle of $G = E/[2(1+\rho)]$02 and a maximum torsion angle of $G = E/[2(1+\rho)]$03. The bending angle-versus-distance curve fluctuated most for $G = E/[2(1+\rho)]$04 mm due to HSV-based vision noise. In torsion, the slope increased notably for $G = E/[2(1+\rho)]$05 mm, with a plateau near $G = E/[2(1+\rho)]$06 attributed to co-acting magnetic forces when the external magnet was too close. ANSYS magneto-mechanical finite-element analysis qualitatively confirmed the intended mode decoupling under the same clamped-base configuration (Zhang et al., 2 Oct 2025).

In phantom intervention, the robot navigated a water-filled PDMS phantom with square channels of width $G = E/[2(1+\rho)]$07 mm, seven routes of length $G = E/[2(1+\rho)]$08–$G = E/[2(1+\rho)]$09 mm, and intersection angles $G = E/[2(1+\rho)]$10–$G = E/[2(1+\rho)]$11. The procedure consisted of lumen following in bending mode, target approach, and then twist-activated drug release. The reported performance highlight is seven successful intervention-release trials. In a representative run, total time from entry to target was $G = E/[2(1+\rho)]$12 s, traversal between intersections was about $G = E/[2(1+\rho)]$13 s, and external-magnet reorientation for branch selection required about $G = E/[2(1+\rho)]$14 s. During release, slight twist appeared at $G = E/[2(1+\rho)]$15 s, pronounced helical deformation without release at $G = E/[2(1+\rho)]$16 s, and tight coiling at $G = E/[2(1+\rho)]$17 s expelled the air segment and drug. A single torsional actuation produced a small dose, while repeated twist/relax cycles drew in surrounding fluid and released progressively diluted doses (Zhang et al., 2 Oct 2025).

Ball-chain MeSCRs provide a complementary performance regime centered on steerability rather than torsional drug release. Under homogeneous $G = E/[2(1+\rho)]$18 mT fields and tip lengths up to $G = E/[2(1+\rho)]$19 mm, the simulated planar workspace area was about $G = E/[2(1+\rho)]$20 mm$G = E/[2(1+\rho)]$21 for the ball chain, compared with about $G = E/[2(1+\rho)]$22 mm$G = E/[2(1+\rho)]$23 for a tip-magnet rod and about $G = E/[2(1+\rho)]$24 mm$G = E/[2(1+\rho)]$25 for a distributed magnetic composite rod, against a theoretical revolute-joint bound of $G = E/[2(1+\rho)]$26 mm$G = E/[2(1+\rho)]$27. Experimental model validation yielded a mean centerline error of $G = E/[2(1+\rho)]$28 mm. In a bifurcating-channel phantom with channels $G = E/[2(1+\rho)]$29 mm wide and $G = E/[2(1+\rho)]$30 mm deep and turning angles $G = E/[2(1+\rho)]$31, $G = E/[2(1+\rho)]$32, $G = E/[2(1+\rho)]$33, $G = E/[2(1+\rho)]$34, and $G = E/[2(1+\rho)]$35, the robot successfully entered all five side branches. Most rotation occurred over two to three spheres at the corner, which explains the small effective bend radius (Pittiglio et al., 2023).

These experiments make clear that MeSCRs are not a single performance point. Radial-tip architectures emphasize modal decoupling, compact mechanism integration, and site-specific release, whereas multi-lumped ball-chain architectures emphasize extreme steerability and small-radius turning. The shared theme is distributed magnetic actuation coupled to a compliant continuum body.

7. Misconceptions, limitations, safety, and research directions

Several recurring misconceptions are clarified by the current literature. First, MeSCRs are not simply axially magnetized magnetic continuum robots with more magnetic material. The axial-magnetization theory proves material-twist-free behavior under stated bounds and coplanar conditions, which explains why many classical designs are effectively bending-only. Radially embedded or otherwise non-axial multi-magnet layouts are specifically introduced to generate torsional authority (Wu et al., 15 Jul 2025). Second, increasing magnetic content does not always improve shape authority: in distributed magnetic composites, adding particles increases both magnetic torque and bending stiffness, whereas ball-chain architectures separate magnetic volume from elastic bending resistance by localizing compliance at inter-sphere contacts (Pittiglio et al., 2023). Third, magnetic force cannot be assumed negligible in every operating regime; although usually smaller than torque, the dual-mode catheter reports that force effects can become visible when the external permanent magnet is too close to the tip (Zhang et al., 2 Oct 2025).

The present MeSCR literature also contains significant reporting gaps. In the dual-mode catheter study, the polymer type, wall thickness, inner and outer diameters, internal magnet grades and remanence, magnet dimensions, spacing along the shaft, and exact elastic moduli $G = E/[2(1+\rho)]$36 and $G = E/[2(1+\rho)]$37 are not reported. Absolute field magnitudes $G = E/[2(1+\rho)]$38 and gradients $G = E/[2(1+\rho)]$39 at the catheter are likewise not quantified. Automated bandwidth and closed-loop accuracy metrics are absent; the manipulation protocol is manual and human-in-the-loop, and operation time is dominated by external-magnet reorientation. These omissions limit precise safety calculations and impede direct reproducibility of magnetic loading conditions (Zhang et al., 2 Oct 2025).

Safety and imaging considerations remain architecture-dependent. No heating issues are reported in the dual-mode catheter because actuation is purely magnetic and uses no onboard electricity. The built-in air segment functions as a safety buffer against premature leakage, and release requires combined distal magnetic torque and proximal twist, reducing the likelihood of accidental release under isolated proximal actuation. At the same time, MRI compatibility is limited by embedded permanent magnets, and fluoroscopy or ultrasound guidance may therefore be preferable for externally magnetized permanent-magnet systems. Clinical constraints include avoiding strong gradients near ferromagnetic implants, preventing interference with devices, and managing exposure to static magnetic fields; the available studies note that these issues must be quantified in future in vivo work (Zhang et al., 2 Oct 2025).

Methodologically, many models remain deliberately simplified. The topology-optimization framework assumes ideal hard-magnetic behavior, neglects demagnetizing fields and internal magnetic interactions, treats the applied field as spatially uniform in its base form, and does not model hysteresis, eddy currents, or heating. Contact is handled through velocity correction, which may locally reduce gradient fidelity even if useful sensitivity information is preserved globally (Wang, 28 Mar 2025). Rod-theoretic reviews identify a parallel gap at the continuum level: magnetic actuation is readily represented as external wrench densities, but explicit standardized magneto-mechanical coupling laws for multi-magnet continua in three-dimensional, spatially varying fields remain an open modeling need, especially when real-time control, contact, and field solvers must be integrated in a single loop (Alessi et al., 2024).

Near-term research directions are correspondingly clear. Reported next steps include automatic mode switching and external-magnet reorientation to reduce operation time; optimization of groove and plate geometries, compliant reservoirs, and flow paths to improve delivery efficiency; larger permanent magnets or electromagnetic arrays to offset the rapid $G = E/[2(1+\rho)]$40 loss of actuation authority with depth; dynamic Cosserat-type models with magneto-elastic coupling; experimental validation of optimal multi-magnet layouts; and integration with imaging and shape sensing (Zhang et al., 2 Oct 2025). Taken together, these directions suggest that MeSCR research is moving from proof-of-principle demonstrations toward quantitatively specified, optimization-driven, and closed-loop controlled systems for minimally invasive intervention.