Navion eMNS: Clinical Magnetic Navigation

• Navion eMNS is a clinical electromagnetic navigation system featuring a three-coil, high-bandwidth actuation array and advanced feedback control for precise magnetic manipulation.
• Its motion-centric control paradigm computes device torque and force directly, reducing coil currents (below 3 A at clinical distances) and expanding the workspace fivefold compared to legacy systems.
• The system employs energy-optimal current allocation and a high-rate (125 Hz) feedback loop to maintain robust actuation within stringent thermal and power limits for surgical applications.

Navion eMNS is a clinically oriented electromagnetic navigation system (eMNS) characterized by a three-coil, high-bandwidth actuation array designed for magnetic manipulation in surgical applications. Unlike legacy multi-coil systems, the Navion adopts system-level control strategies to substantially expand the effective workspace for real-time, torque-controlled navigation of magnetic devices. Its distinctive architecture, energy-optimal current allocation, and feedback-driven control enable robust operation within stringent clinical thermal and power limits at significantly greater distances from the actuation coils (Zughaibi et al., 23 Nov 2025).

1. System Architecture and Electromagnetic Model

The Navion eMNS consists of three identical circular coils configured in a horizontal equilateral triangle (120°–120°–120°) around the workspace center. The coil centers reside in a horizontal plane, each axis slightly tilted upward, resulting in overlapping magnetic fields and a quasi-spherical actuation region several centimeters in diameter. Independent commercial drivers (Elmo Gold Drum HV) power each coil, with a clinical current limit of ±25 A per channel for thermal safety, although hardware is rated up to ±45 A (Zughaibi et al., 23 Nov 2025).

The magnetic field at position $p$ in the workspace is modeled linearly as:

$$\mathbf{b}(p) = \sum_{i=1}^3 I_i\,\mathbf{b}_i(p) = \mathbf{U}_b(p)\,\mathbf{I}$$

where $I_i$ is the current in coil $i$, $\mathbf{b}_i(p)$ is its precomputed unit-current field map (via multipole expansion and calibration), $\mathbf{U}_b(p) \in \mathbb{R}^{3 \times 3}$ aggregates these maps, and $\mathbf{I} = [I_1, I_2, I_3]^T$. For force generation, the spatial magnetic field gradient is similarly linear in current:

$$\nabla\mathbf{b}(p) = \mathbf{U}_G(p)\,\mathbf{I}, \quad \mathbf{U}_G(p) \in \mathbb{R}^{5 \times 3}$$

Thermal and bandwidth constraints (driver: 24.5 Hz at 5 A; command update: 125 Hz) directly bound the actuation workspace—for distant targets, the required current exceeds safe operational levels (Zughaibi et al., 23 Nov 2025).

2. Control Paradigms: Motion-Centric Versus Field-Centric

Traditional field-centric (open-loop) eMNS operation commands a target field direction and magnitude, placing the burden of mechanical actuation on the alignment between the magnetic dipole and the field. This approach results in wasted energy in field components parallel to the dipole that produce negligible mechanical work. In contrast, Navion implements a motion-centric paradigm: the feedback controller directly computes the desired torque $\boldsymbol{\tau}_c$ and, if necessary, force $\mathbf{f}_c$ for the device, ensuring all generated fields contribute to the intended mechanical actuation (Zughaibi et al., 23 Nov 2025).

The mapping from device motion objectives to coil currents leverages a Jacobian/allocation matrix,

$$\boldsymbol{\tau} = \mathbf{m} \times \mathbf{b} = \mathrm{skew}(\mathbf{m})\,\mathbf{U}_b(p)\,I = J(p, \alpha, \beta)\,I$$

where $J(p,\alpha,\beta) \in \mathbb{R}^{3\times3}$ incorporates the device dipole, its orientation $(\alpha,\beta)$, and the spatially-varying coil field derivatives.

3. Feedback Architecture and Real-Time Optimization

A high-rate (125 Hz) clinical-grade optical tracking system measures the 3D position and orientation of the dipole, synchronizing with coil-driver command updates. The control loop centers on a decoupled Linear Quadratic Regulator with integral action (LQRI), which computes the torque setpoints $\tau_{c,x}, \tau_{c,y}$ for closed-loop stabilization tasks such as inverted pendulum balancing (Zughaibi et al., 23 Nov 2025).

At each cycle, coil currents are set by solving the quadratic program:

$$i^\star = \arg\min_{i \in \mathbb{R}^3} \frac{1}{2}\|i\|_2^2 \quad \text{subject to} \quad J(p, \alpha, \beta)\,i=\tau_c,\quad \|i\|_\infty \leq 25\,\mathrm{A}$$

The unique energy-minimizing solution is provided by the Moore–Penrose pseudoinverse:

$i^\star = J^\dagger(p,\alpha,\beta)\,\tau_c$

This approach ensures that generated torques and forces are attained with the minimal possible current, judiciously exploiting redundancy in the triangular coil geometry (Zughaibi et al., 23 Nov 2025).

4. Workspace Expansion, Actuation Redundancy, and Performance

The use of motion-centric objectives coupled with feedback-based, energy-optimal current allocation markedly reduces current draw: pendulum stabilization on OctoMag required 8–14 A per coil (field-alignment) but just 0.1–0.2 A per coil (motion-centric), while Navion maintained balancing below 3 A per coil at clinical distances. Critically, field-alignment control limited the achievable workspace to ~10 cm radius, whereas the feedback-optimized Navion system preserved performance out to at least 50 cm—a fivefold increase in radius and approximately 125-fold increase in workspace volume (Zughaibi et al., 23 Nov 2025).

The “Feasibility Margin” at position $p$,

$$\mathrm{FM}(p) = 25\,\mathrm{A} - \|J^\dagger(p,\alpha,\beta)\|_\infty$$

remains positive across this enlarged workspace, confirming current saturation is avoided during dynamic tasks.

Key factors enabling this workspace expansion are summarized as follows:

Design Feature	Impact on Workspace	Mechanism
Motion-centric torque/force objectives	Eliminates wasted field directions	Directs energy to task
Energy-optimal allocation	Reduces required current	Minimizes $\|\mathbf{I}\|$
High-bandwidth/real-time feedback	Enables faithful torque application	Adapts at 125 Hz
Triangular coil redundancy	Exploits spatial inhomogeneities	Allocates current locally

Together, these allow stable and robust device control significantly farther from the coil surface, with no hardware change other than software and control adaptation (Zughaibi et al., 23 Nov 2025).

5. Multi-Agent Control and System Generalizability

Although multi-agent dynamic balancing demonstrations were realized on an eight-coil OctoMag platform, the Navion architecture theoretically supports multi-agent scenarios. With three coils, two identical dipoles can be balanced if both reside along a common radial line and control objectives are decoupled. This leverages both coil redundancy and nonlinear spatial decay of field strength to enable independent actuation despite minimal hardware count. The control laws and current allocation strategies are directly transferrable to other clinical eMNS arrays, provided coil calibration and actuation maps are available (Zughaibi et al., 23 Nov 2025).

6. Implications for Modeling and Machine Learning Integration

Accurate field modeling is pivotal for eMNS control. While the foundational Navion implementation relies on linear multipole expansions for $\mathbf{U}_b(p)$, recent work on similar eMNS arrays (e.g., CardioMag’s eight-coil system) demonstrates that nonlinearities arising from ferromagnetic core saturation and coil coupling degrade linear model accuracy above 15 A. Machine learning models, notably artificial neural networks (ANNs), reduce the root mean squared error (RMSE) of field magnitude by ~80% over linear multipole models and achieve sub-5 mT error even at 35 A. This improved model fidelity directly impacts forward/inverse kinematics, enhancing device localization and targeting accuracy at high fields—a critical concern for Navion-class clinical applications (Yu et al., 2019).

Ongoing challenges include ensuring model generalization outside the calibration data grid, integrating low-latency inference and online adaptation to thermal drift, and extension to dynamic actuation scenarios. Incorporating ANN-based field models into the control loop offers the potential for further robustness and reliability in surgical navigation (Yu et al., 2019).

7. Clinical Relevance and Future Directions

The Navion eMNS advances the scalability and efficiency of magnetic manipulation for surgery by bridging device-level actuation requirements and hardware-level constraints through feedback-driven, energy-efficient control. Its ability to expand the actuation workspace without hardware redesign underscores the importance of dynamic feedback, optimized current allocation, and real-time tracking in contemporary clinical eMNS. Future directions include field model integration via machine learning, exploration of multi-agent actuation in more general spatial configurations, and the realization of robust, lightweight firmware for low-latency, closed-loop device navigation in vivo (Zughaibi et al., 23 Nov 2025, Yu et al., 2019).