MagServo: Uncertainty-Resilient Hierarchical Magnetic Servoing via Learned Latent Representations

Yuhan Tan; Yameng Zhang; Pei Liu; Yao Zhong; Zhongliang Jiang

cs.ROOct 5, 2026

MagServo: Uncertainty-Resilient Hierarchical Magnetic Servoing via Learned Latent Representations

Authors: Yuhan Tan, Yameng Zhang, Pei Liu, Yao Zhong, Zhongliang Jiang

Organizations: Department of Bioengineering, Imperial College London, UK. · Department of Mechanical Engineering, The University of Hong Kong, Hong Kong SAR. · School of Computation, Information and Technology, Technical University of Munich, Germany.

Abstract

Magnetic navigation provides contact-free and line-of-sight-independent feedback for robotic systems, yet existing approaches typically rely on explicit pose estimation or direct use of raw magnetic measurements, making accurate control susceptible to modeling errors, measurement noise, and disturbances. This work presents MagServo, a hierarchical learning-based framework for robust 6-DoF magnetic servoing directly using the learned latent magnetic feature. MagServo learns uncertainty-resilient magnetic representations through masked reconstruction and captures state-dependent interaction dynamics between robot motion and latent magnetic transitions without analytical magnetic models or explicit Jacobian supervision. Based on the learned dynamics, a hierarchical controller combines nonlinear model predictive control for coarse approach with local Jacobian inversion for precise fine regulation. Extensive physical experiments demonstrate submillimeter and subdegree accuracy, achieving mean terminal errors of 0.386 mm and 0.479 degree for 6-DoF pose reaching. MagServo further outperforms a localization-based control baseline in complex trajectory tracking and maintains robust performance under unseen magnetic-source configurations without retraining. A supplementary video of the real-robot experiments is available at https://youtu.be/rZt1NUP1Mr0.

Figures & tables

Fig. 1: Magnetic servoing system comprising a robot manipulator, a dual-magnet end-effector fixture, and a $4\times 4$ magnetometer array.

Fig. 2: Three-stage magnetic representation training. Stage I learns magnetic representations by reconstructing complete observations from masked inputs. Stage II learns latent dynamics with the encoder frozen and selects representative anchors through clustering. Stage III jointly refines the representation and dynamics using reconstruction and prediction losses, with Rosetta anchor regularization to limit latent coordinate drift.

Fig. 3: Closed-loop control architecture of MagServo. A shared encoder maps target and measured magnetic fields to latent states. The latent-error norm selects between coarse MPC and local Jacobian inversion. The selected action $\mathbf{u}$ drives the robot, closing the loop through magnetic feedback.

Fig. 4: Single-pose reaching results. (a) Servoing trajectories from 12 different initial poses toward the same target pose. (b) Heatmaps of the $4\times 4$ magnetic measurements at iterations 0, 30, 60, and 90 compared with the target reference. The table reports the corresponding tracking errors and the final errors at the termination step.

Trajectory	$e_{p}$ (mm)			$e_{r}$ ( ∘ )			Actions
	$\delta_{x}$	$\delta_{y}$	$\delta_{z}$	$\phi_{x}$	$\phi_{y}$	$\phi_{z}$
Spiral	$0.179\pm 0.196$	$0.238\pm 0.223$	$0.055\pm 0.048$	$0.156\pm 0.136$	$0.128\pm 0.122$	$0.084\pm 0.069$	673
Raster	$0.255\pm 0.311$	$0.580\pm 0.485$	$0.118\pm 0.120$	$0.673\pm 0.440$	$0.288\pm 0.343$	$0.235\pm 0.239$	894

TABLE I: Tracking accuracy for the selected spiral and raster trajectories.

Fig. 5: Trajectory-following results. (a) Spiral and (b) raster trajectories. The left column shows the reference and executed servoing trajectories, while the middle and right columns show the corresponding position and rotation errors, respectively.

Trajectory	Method	$e_{p}$ (mm)	$e_{r}$ ( ∘ )
Spiral	Localization	$1.409\pm 0.673$	$1.559\pm 0.627$
	MagServo	$\mathbf{0.333\pm 0.267}$	$\mathbf{0.249\pm 0.155}$
Raster	Localization	$1.650\pm 0.795$	$1.538\pm 0.589$
	MagServo	$\mathbf{0.712\pm 0.505}$	$\mathbf{0.855\pm 0.478}$

TABLE II: MagServo versus the localization-based baseline.

Trajectory	Magnetic Sources	$e_{p}$ (mm)			$e_{r}$ ( ∘ )			Actions
		$\delta_{x}$	$\delta_{y}$	$\delta_{z}$	$\phi_{x}$	$\phi_{y}$	$\phi_{z}$
Spiral	Two	$0.179\pm 0.196$	$0.238\pm 0.223$	$0.055\pm 0.048$	$0.156\pm 0.136$	$0.128\pm 0.122$	$0.084\pm 0.069$	673
	Three	$0.304\pm 0.245$	$0.268\pm 0.248$	$0.119\pm 0.093$	$0.253\pm 0.176$	$0.292\pm 0.202$	$0.122\pm 0.093$	684
Raster	Two	$0.255\pm 0.311$	$0.580\pm 0.485$	$0.118\pm 0.120$	$0.673\pm 0.440$	$0.288\pm 0.343$	$0.235\pm 0.239$	894
	Three	$0.239\pm 0.289$	$0.491\pm 0.489$	$0.123\pm 0.122$	$0.408\pm 0.364$	$0.226\pm 0.285$	$0.208\pm 0.223$	824

TABLE III: Tracking accuracy with two and three magnetic sources.

Explore similar work

Magnet-Aware Control of Legged Robots
Oct 6, 2026J. Playan Garai, S. B. Djuve, C. McGreavy +1Legged Robots
Technical University of Munich (TUM), Munich, Germany · Conseil Europ´een pour la Recherche Nucl´eaire (CERN), BE/CEM/MRO, Geneva, Switzerland
From Transportation to Manipulation: Enabling Grasping in Magnetic Robotics
Sep 14, 2026Lara Bergmann, Noah Greis, Cedric Grothues +2Gripper Robotic Manipulator
Faculty of Technology, CITEC, Bielefeld University, Germany · Fraunhofer IOSB-INA, Lemgo, Germany
A Unified Benchmark for RCM-Constrained Visual Servoing: Modeling-Controller Interaction and Robustness Analysis in Laparoscopic Robots
Jun 22, 2026Jing Zhang, Mengtang LiVisual Servoing Surgical Robots
School of Intelligent Systems Engineering, Shenzhen Campus of Sun Yat-sen University, Shenzhen, China

Oct 6, 2026, cs.ROJ/K move · Enter open · S save

Magnet-Aware Control of Legged Robots

J. Playan Garai, S. B. Djuve, C. McGreavy, M. Khadiv

Technical University of Munich (TUM), Munich, Germany · Conseil Europ´een pour la Recherche Nucl´eaire (CERN), BE/CEM/MRO, Geneva, Switzerland

Legged Robots

Open paper AskPDF arXiv

Autonomous robots can increase uptime and reduce human exposure in Big Science facilities, but strong magnetic fields needed for their operation corrupt sensors and induce pose-dependent mechanical wrenches that destabilize robots and challenge conventional reactive controllers. This paper presents a control framework for modeling, estimating, and dynamically compensating for spatially varying magnetic wrenches acting on legged robots to improve robustness in these fields. We introduce a custom physics plugin for the MuJoCo simulator to model magnetic forces on rigid-body elements, alongside an inverse field-estimation framework to infer the latent magnetic field directly from quadruped dynamic responses and any number of sensor readings. Furthermore, we develop a Magnet-Aware Model Predictive Control (MPC) and Whole-Body Control (WBC) architecture that predicts and counteracts magnetic perturbations during locomotion to increase the range of magnetic fields in which the robot can operate. The effectiveness of the framework is validated through both simulation and physical hardware experiments. We show our method increases the the maximum rejectable disturbances from magnetic field in the force space by a factor of 2.5 and and between 1.6-2 times in torque space compared to a non-compensated system. Within the proposed magnetic field, this constitutes an increase in the area in which the robot can operate by 29,34% of which 10,84% would have previously caused an immediate collapse to non-compensated controllers.

Abstract

Explore similar work

Magnet-Aware Control of Legged Robots

From Transportation to Manipulation: Enabling Grasping in Magnetic Robotics

A Unified Benchmark for RCM-Constrained Visual Servoing: Modeling-Controller Interaction and Robustness Analysis in Laparoscopic Robots