Electromagnetic Micro-Guidewire Control in Large Workspaces

• Small magnetic volumes and inverse-cube field decay make sufficient actuation at clinically relevant distances challenging under coil-current and thermal limits.The paper notes that workspace depends on more than hardware alone: real-time pose feedback previously enabled energy-efficient allocation and expanded operational workspace by up to an order of magnitude [7].Source: §I · p. 1
• Pose-dependent torque allocation exploits nonunique torque generation: fields differing in their component parallel to the tip’s magnetic moment produce identical torques but can require different electrical effort.Real-time pose feedback is needed because the current-to-torque mapping depends on tip position and orientation [7].Source: §I · p. 1
• Magnetic guidewires apply steering torque at the distal end, bypassing torsional transmission along the shaft and enabling remote tip orientation during advancement,.Computer-controlled actuation also supports teleoperation and precise tip positioning.Source: §I · p. 2
• The study evaluates pose-feedback efficiency at 15 Hz, representative of clinical fluoroscopy, and identifies fluoroscopic pose reconstruction and registration as future work.Passive inductive pick-up coils are also proposed as a possible route to continuous, non-ionizing pose feedback.Source: §I · p. 2
• 43.7 cm α-plane and 54.9 cm β-plane tracking from the coil cover achieved angular RMSE below 0.25° despite active current constraints.The framework combines pose-dependent convex torque allocation, dynamic feedback, and repetitive control.Source: §I · p. 2

A. Micro-Guidewire

• The 0.6 mm-diameter tubular permanent magnet is mounted at the distal tip of a 100 µm superelastic NiTi base wire; its magnetic-material volume is 2.47 mm3.The applied field typically remains below 10 mT, substantially below μ0Hc ≈90.5 mT, so field-induced magnetization changes are neglected and the dipole moment is assumed constant in the magnet-fixed frame.Source: A. Micro-Guidewire · p. 2

B. Tracking, Motorized Advancer, and Anatomical Model

• Two synchronized, approximately orthogonal monochrome-camera views support segmentation, principal-component axis detection, and stereo reconstruction of magnet-center position and longitudinal-axis direction.Motion-capture markers register the holder frame to the eMNS frame; images are acquired at 100 Hz or 15 Hz, the latter representing clinical fluoroscopy.Source: B. Tracking, Motorized Advancer, and… · pp. 2–3
• A joystick-controlled motorized advancer inserts or retracts the guidewire in a patient-specific in-vitro model of the cerebral ventricular system.The motor sets advancement direction and speed while magnetic actuation controls distal-tip orientation.Source: B. Tracking, Motorized Advancer, and… · p. 3

C. Electromagnetic Navigation System

• The three lateral coils are independently driven, and the eMNS supports up to 45 A per coil with current references updated at 100 Hz.Its identified electrical bandwidth is 24.5 Hz for a 5 A current-reference amplitude.Source: C. Electromagnetic Navigation System · p. 3

III. MODELING

• The coupled eMNS–guidewire dynamics inform achievable feedback performance, while the magnetic actuation model relates coil currents to effective tip torque for pose-dependent allocation.Source: §III · p. 3

A. Identification of the Guidewire Dynamics

• The guidewire dynamics are identified from commanded magnetic torque to measured tip angle in air and water, assessing closed-loop limits and the frequency range for tracking and disturbance rejection.Periodic multisines use selected frequencies from 0–10 Hz; randomized phase realizations support best-linear-response estimation and reveal nonlinear distortion when total uncertainty substantially exceeds the stochastic-noise estimate.Source: A. Identification of the Guidewire… · p. 3
• Approximately 4 Hz is the combined-system −3 dB bandwidth in water.The identified responses are approximated by second-order models with effective delay; the result suggests potential for active compensation of physiological disturbances.Source: A. Identification of the Guidewire… · p. 3

B. Magnetic Actuation Model

• At each tip position, coil currents map linearly to magnetic field and its spatial gradient, with position dependence modeled by a quadrupole-truncated multipole expansion.The model assumes linear dependence on current; deviations from linear superposition may occur above 40 A.Source: B. Magnetic Actuation Model · pp. 3–4
• In the β direction, fitted natural frequency f0 is (4.8, 2.9) Hz and damping D is (0.035, 0.40) in air and water, respectively.The caption also reports static gain K = (9.8, 8.3) × 10^5 rad/(N m) and delay Td = (48, 15) ms; the dynamics, especially K, may vary with guidewire configuration.Source: Figure 2 · p. 4
• The effective control torque combines direct magnetic torque with the moment from magnetic force about a local center of rotation.The force-induced moment is approximated using a lever arm r ≈ ℓn; the controllable torque has two components perpendicular to the dipole axis.Source: B. Magnetic Actuation Model · p. 4

IV. CONVEX CURRENT ALLOCATION UNDER CONSTRAINTS

• When pseudoinverse currents exceed bounds, a convex regularized least-squares allocation minimizes torque error subject to per-coil current limits.As regularization approaches zero, the solution converges to the minimum-2-norm current among bounded vectors minimizing torque error; reported experimental residuals were of order 10^-6–10^-5.Source: §IV · p. 5
• Nullspace redistribution extends the current-limited workspace beyond the unconstrained pseudoinverse allocation.The approach exploits the nonuniqueness of current realizations to manage coil loading under current bounds.Source: §IV · p. 5
• Nullspace currents redistribute coil loading without changing commanded torque, and the pose-dependent nullspace requires real-time pose feedback.At regular poses the allocation has a one-dimensional nullspace; further rank loss is a magnetic actuation singularity.Source: §IV · p. 5
• At 437 mm, optimized current redistribution maintained tracking with a 7×10^-6 torque residual, while componentwise clipping caused 6%–8% torque errors.Comparable performance was observed at 100 Hz and 15 Hz for the tested trajectory.Source: §IV · p. 5
• At 549 mm, β-angle tracking with 5.5° amplitude had 0.24° RMS error despite an active current constraint.The relative torque residual was 1.3 × 10^-6; β motion has greater coil torque authority than α motion parallel to the coil cover.Source: §IV · p. 5

V. CONTROL

• Automatic reduced-attitude regulation with repetitive control establishes reproducible comparisons of field-alignment and torque-based allocation at matched tip positions and angular trajectories.The experiments also demonstrate achievable tracking accuracy.Source: §V · pp. 5–6
• Approximately 7·10−6 relative torque residual under active current constraints compares with 6 %-8 % for componentwise pseudoinverse clipping.Fig. 4 shows constrained torque allocation during tip-angle tracking at 437 mm from the coil cover.Source: Figure 4 · p. 6
• Pose regulation alone does not constrain vessel-wall interaction forces in uncertain, contact-rich environments.Hybrid pose/force control and compliance control are potential extensions.Source: §V · p. 6

A. Manual Torque Control

• The operator commands two torque components using a joystick and adjusts them by visual observation, while allocation uses measured tip position and dipole direction.No automatic attitude-error feedback is applied in this mode.Source: A. Manual Torque Control · p. 6
• At 549 mm from the coil cover, active current constraints coincide with nullspace redistribution, a 77° field–dipole angle, and a 1.3 · 10−6 relative torque residual.Fig. 5 depicts tracking with a 45 A current limit.Source: Figure 5 · p. 6
• Coordinated advancement and torque commands enable traversal of more of the vascular model, with bending torques reaching 10 µNm.The supplementary video shows this mode with the motorized advancer.Source: A. Manual Torque Control · p. 6

B. Reduced Attitude Feedback Control

• Rotation about the dipole axis is uncontrollable, so the controlled reduced attitude is its direction n on the unit sphere S2.Both magnetic torque and force-induced moment are orthogonal to n under the effective-torque model.Source: B. Reduced Attitude Feedback Control · p. 6
• The proportional-derivative feedback law is formulated directly in reduced attitude to avoid singularities associated with local angular parameterizations.The body-fixed angular velocity and proportional and damping gains are identified in the formulation.Source: B. Reduced Attitude Feedback Control · pp. 6–7
• In water, identified damping D = 0.40 led the experiments to set Kd = 0; in air, derivative feedback and a delay-aware Kalman filter provided additional damping and mitigated phase-margin loss.The air experiments are presented in the supplementary video.Source: B. Reduced Attitude Feedback Control · p. 7

C. Repetitive Control

• Periodic reference trajectories and repeatable plant dynamics produce recurring errors, which the scheme addresses by learning a periodic feedforward torque correction.A separate correction vector is stored for each controlled angle.Source: C. Repetitive Control · p. 7
• A phase-lead update uses a shift of about 65 ms to approximate measured closed-loop lag and accumulates errors across repetitions as iteration-domain integral action.The shift associates the correction with a phase preceding the observed error.Source: C. Repetitive Control · p. 7
• After each cycle, a centered 17-sample moving-average Q-filter smooths the correction without time delay and attenuates high-frequency components to limit learning near guidewire resonance.The filter wraps around cycle boundaries.Source: C. Repetitive Control · p. 7

D. Field-Alignment Benchmark

• Field magnitude and controller gain are paired to match local magnetic stiffness for the benchmark.The commanded field angle is uφ.Source: D. Field-Alignment Benchmark · p. 7
• The field-alignment controller learns a field-direction correction while retaining prescribed field magnitude, whereas torque-based allocation uses available dipole-orientation information directly.The prescribed field is allocated using a position-dependent pseudoinverse; orientation feedback is noted as useful for torque-based allocation.Source: D. Field-Alignment Benchmark · p. 7
• Torque-based allocation achieves comparable angular tracking with substantially lower current demand than field alignment.The comparison uses identical angular references and magnet positions.Source: D. Field-Alignment Benchmark · p. 7

VI. DISCUSSION AND PERSPECTIVE

• Pose-informed torque allocation reduces current demand, while optimized coil-current redistribution maintains accurate torque generation at individual current limits.Combined with feedback and repetitive control, the framework demonstrates sub-degree tracking up to 55 cm for a 2.47 mm3 tip-magnet volume in realistic anatomical models.Source: §VI · p. 7
• Clinical translation requires reliable in-vivo state estimation and registration to the electromagnetic navigation system.The efficiency benefit persists at 15 Hz, motivating investigation of X-ray pose feedback and complementary non-ionizing in-vivo localization.Source: §VI · p. 7
• Operator guidance, controlled quantities, and feedback-algorithm design remain open questions for in-vivo applications.The authors identify these as topics for future exploration alongside partially automated navigation and physiological-disturbance rejection.Source: §VI · p. 7
• The authors report occasional camera-measurement artifacts, attributed primarily to air bubbles and optical refraction through the vascular model.The figure compares matched-stiffness tip-angle tracking and current norm for field alignment and torque-based control.Source: Figure 6 · p. 8

CONFLICT OF INTERESTS

• Bradley Nelson is disclosed as a co-founder of MagnebotiX AG and Nanoflex Robotics AG; the other authors declare no conflict of interest.Source: CONFLICT OF INTERESTS · p. 8