pith. sign in

arxiv: 2505.16662 · v4 · submitted 2025-05-22 · 💻 cs.RO · eess.SP

Joint Magnetometer-IMU Calibration via Maximum A Posteriori Estimation

Chuan Huang , Gustaf Hendeby , Isaac Skog This is my paper

Pith reviewed 2026-05-22 13:37 UTC · model grok-4.3

classification 💻 cs.RO eess.SP
keywords magnetometer calibrationIMU calibrationjoint calibrationmaximum a posteriori estimationsensor fusioninertial navigationorientation estimation
0
0 comments X

The pith

A maximum a posteriori method jointly calibrates magnetometer and IMU pairs with 20-30% lower root mean square error than prior approaches while remaining computationally competitive.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a joint calibration technique that simultaneously estimates both the unknown calibration parameters for a magnetometer and IMU and the sensors' orientation trajectory. By casting the problem as maximum a posteriori estimation and supplying analytically derived derivatives, the optimizer can refine the parameters efficiently without relying on numerical gradients. Simulation trials show this yields the lowest root mean square error among compared methods, translating to a 20-30% accuracy improvement. Real-world tests further demonstrate that thirty sensor pairs can be calibrated in under two minutes on ordinary hardware and that the resulting calibration supports magnetic-field-aided inertial navigation at performance levels matching slower but highly accurate alternatives.

Core claim

Treating calibration parameters and orientation trajectory as unknowns in a single maximum a posteriori estimation problem, with analytically derived derivatives for efficient optimization, produces calibration parameters whose root mean square error is 20-30% lower than that of two state-of-the-art methods while preserving competitive run times; the same calibration also enables inertial navigation whose positioning accuracy is comparable to that obtained with the most accurate reference method.

What carries the argument

Maximum a posteriori estimation framework that jointly optimizes calibration parameters and the full orientation trajectory using analytically derived derivatives.

If this is right

  • Calibrated magnetometer-IMU pairs can be produced rapidly enough for routine field recalibration on consumer laptops.
  • Magnetic-field-aided inertial navigation systems achieve positioning performance comparable to that obtained with slower reference calibrations.
  • The analytic derivatives reduce the computational burden of the joint optimization relative to purely numerical approaches.
  • Thirty sensor pairs can be processed in under two minutes, enabling batch calibration workflows that were previously impractical.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same joint MAP structure could be extended to other sensor triads that share a common rigid frame and require orientation estimation.
  • Because the derivatives are analytic, the method may integrate directly into real-time estimators that must occasionally re-calibrate without restarting from scratch.
  • Faster calibration cycles could allow periodic in-situ recalibration during long-duration robot missions, mitigating drift from temperature or aging effects.

Load-bearing premise

The reported accuracy and speed advantages hold only if the two comparison methods were implemented and tuned to the same standard and if the chosen simulation and real-world conditions represent typical magnetometer-IMU operating environments.

What would settle it

An independent replication that applies all three algorithms to the same new set of magnetometer-IMU recordings with different motion patterns and reports whether the proposed method still exhibits the lowest root mean square calibration error.

Figures

Figures reproduced from arXiv: 2505.16662 by Chuan Huang, Gustaf Hendeby, Isaac Skog.

Figure 1
Figure 1. Figure 1: The reference (big axes sets) and sensors’ coordinate frames (small axes sets). The Z-axis in the reference frame is aligned with the local gravity vector, g, the Y–Z plane is parallel to the local magnetic field, m(α), with its horizontal component pointing in the positive Y-axis direction. The dip angle of the magnetic field α is the angle between the magnetic field and the horizontal plane. The misalign… view at source ↗
Figure 2
Figure 2. Figure 2: The benefit of this particular choice is that the orientation can be fully described using only the roll (ϕ) and pitch (γ) angles. This is because yaw (ψ) represents a rotation around the magnetic field direction, which does not change the relative inclination of the sensor to the field. Let Rm k be defined in terms of roll and pitch angles, i.e., R m k ≜ R m k (ϕk, γk) = Rx(ϕk)Ry(γk) (20a) where Rx(ϕk) = … view at source ↗
Figure 3
Figure 3. Figure 3: a that the computation time of all three methods in￾creases almost linearly with the sampling frequency, which implies that the processing time increases linearly with the number of sensor measurements. This is consistent with the computational complexity upper bounds discussed in Section III-C. Furthermore, the method by Wu et al. is the fastest, followed by the proposed method and the method by Kok et al… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of the computation time (a) and RMSE of the estimated calibration parameters (b - e) on datasets with different sampling rate ratios. The horizontal axis is the ratio of the IMU sampling rate to the magnetometer sampling rate. To evaluate the proposed method’s sensitivity to the initial value of the gyroscope bias, the proposed algo￾rithm is initialized with different bias values perturbed aroun… view at source ↗
Figure 5
Figure 5. Figure 5: The sensor board used in the experiment. It has 30 PNI RM3100 magnetometers and an Osmium MIMU 4844 IMU mounted on the bottom side. the calibration parameters and orientation trajectory. The proposed method was compared with two state-of-the-art methods in terms of computational efficiency and cali￾bration accuracy. Simulation results showed that the pro￾posed method achieves lower RMSE in calibration para… view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison on real-world data. (a) The compu￾tation time of the calibration algorithms. (b) The error-to-traveled￾distance ratios of a magnetic field-aided inertial navigation system that uses the uncalibrated and calibrated magnetometer data. The labels on the x-axis represent different datasets. (c) The horizontal trajectories (LP-2) estimated by the magnetic field-aided inertial navigation s… view at source ↗
read the original abstract

This paper presents a new method for jointly calibrating a magnetometer and inertial measurement unit (IMU), focusing on balancing calibration accuracy and computational efficiency. The proposed method is based on a maximum a posteriori estimation framework, treating both the calibration parameters and orientation trajectory of the sensors as unknowns. This method enables efficient optimization of the calibration parameters using analytically derived derivatives. The performance of the proposed method is compared against that of two state-of-the-art methods. Simulation results demonstrate that the proposed method achieves the lowest root mean square error in calibration parameters, increasing the calibration accuracy by 20-30%, while maintaining competitive computational efficiency. Further validation through real-world experiments confirms the practical benefits of the proposed method. The proposed method calibrated 30 magnetometer-IMU pairs in under two minutes on a consumer-grade laptop, which is one order of magnitude faster than the most accurate state-of-the-art algorithm as implemented in this work. Moreover, when calibrated using the proposed method, a magnetic-field-aided inertial navigation system achieved positioning performance comparable to when it is calibrated with the state-of-the-art method. These results demonstrate that the proposed method is a reliable and effective choice for jointly calibrating magnetometer-IMU pairs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a maximum a posteriori (MAP) estimation framework for joint magnetometer-IMU calibration in which both the calibration parameters and the sensor orientation trajectory are treated as unknowns. Analytically derived derivatives are used to enable efficient optimization of the calibration parameters. The method is compared to two state-of-the-art algorithms; simulation results claim the lowest RMSE in calibration parameters with a 20-30% accuracy gain while remaining computationally competitive. Real-world experiments report that 30 magnetometer-IMU pairs can be calibrated in under two minutes on a consumer laptop (one order of magnitude faster than the most accurate baseline as implemented) and that the resulting calibration yields comparable positioning performance in a magnetic-field-aided inertial navigation system.

Significance. If the reported accuracy and speed advantages prove robust under fair and reproducible experimental conditions, the work would offer a practical advance for batch calibration tasks common in robotics and navigation. The combination of joint MAP estimation with closed-form derivatives addresses a useful efficiency-accuracy trade-off that existing methods appear to handle less favorably.

major comments (2)
  1. [Simulation results] Simulation results section: the central claim of a 20-30% RMSE reduction is load-bearing for the paper's contribution, yet the manuscript provides no quantitative information on hyper-parameter tuning effort, initial conditions, or convergence criteria applied to the two baseline algorithms. Without this, it is impossible to determine whether the observed margin arises from the MAP formulation or from unequal implementation quality.
  2. [Real-world experiments] Real-world experiments: the statement that the proposed method is 'one order of magnitude faster' than the most accurate SOTA algorithm as implemented requires explicit reporting of the optimization library, stopping tolerances, and hardware used for all three methods. The current description leaves open the possibility that the speed advantage is an artifact of differing implementation choices rather than an intrinsic property of the estimator.
minor comments (2)
  1. [Abstract] The abstract and results text interchangeably use 'increasing the calibration accuracy by 20-30%' and 'lowest root mean square error'; a single consistent metric (e.g., percentage reduction in RMSE) would improve clarity.
  2. [Figures] Figure captions for the real-world navigation trajectories should include the number of trials and the precise definition of the positioning error metric to allow direct comparison with the simulation results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We address each major comment point by point below and will revise the paper to enhance reproducibility and transparency while preserving the core contributions.

read point-by-point responses

  1. Referee: [Simulation results] Simulation results section: the central claim of a 20-30% RMSE reduction is load-bearing for the paper's contribution, yet the manuscript provides no quantitative information on hyper-parameter tuning effort, initial conditions, or convergence criteria applied to the two baseline algorithms. Without this, it is impossible to determine whether the observed margin arises from the MAP formulation or from unequal implementation quality.

    Authors: We agree that additional quantitative details on the baseline implementations are necessary to substantiate the accuracy claims and rule out implementation artifacts. In our experiments, the baselines were re-implemented according to their original publications, with initial conditions generated from the same attitude initialization procedure used for the proposed method and convergence enforced when the gradient norm fell below 1e-8. Hyperparameters were tuned via a modest grid search over regularization coefficients to minimize RMSE on the simulated data. In the revised manuscript we will insert a new subsection under Simulation Results that tabulates the exact hyper-parameter values, initial guesses, and stopping criteria applied to each baseline, thereby confirming that the reported 20-30% RMSE improvement is attributable to the MAP formulation. revision: yes

  2. Referee: [Real-world experiments] Real-world experiments: the statement that the proposed method is 'one order of magnitude faster' than the most accurate SOTA algorithm as implemented requires explicit reporting of the optimization library, stopping tolerances, and hardware used for all three methods. The current description leaves open the possibility that the speed advantage is an artifact of differing implementation choices rather than an intrinsic property of the estimator.

    Authors: We acknowledge that explicit reporting of the computational setup is required to support the efficiency claims. All three methods were coded in Python 3.9 and optimized with SciPy's minimize routine (L-BFGS-B solver) on an identical consumer laptop (Intel Core i7-10700K, 32 GB RAM). A uniform stopping tolerance of 1e-10 in function-value change or a maximum of 1000 iterations was applied across methods. The revised real-world experiments section will contain a table listing the optimization library, solver settings, hardware specifications, and measured wall-clock times for each algorithm, demonstrating that the observed speed-up arises from the closed-form derivatives rather than disparate implementation choices. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation is self-contained MAP optimization benchmarked externally

full rationale

The paper introduces a MAP-based joint calibration estimator with analytically derived derivatives for optimization. It directly compares performance against two independent state-of-the-art algorithms on both simulated trajectories and separate real-world experiments, reporting RMSE improvements and runtime metrics without any load-bearing self-citations, fitted-input renamings, or self-definitional steps. The central claims rest on external benchmarks and independent validation data rather than reducing to the method's own inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Based on the abstract alone, the method rests on standard probabilistic modeling assumptions for sensor errors and the existence of a well-behaved optimization landscape for the joint estimation problem. No new entities are introduced.

free parameters (1)
  • calibration parameters
    Biases, scale factors, and misalignment terms are estimated as unknowns within the MAP objective.
axioms (1)
  • domain assumption Sensor measurement errors follow the probabilistic distributions assumed in the MAP formulation.
    Standard modeling choice in IMU and magnetometer calibration literature.

pith-pipeline@v0.9.0 · 5740 in / 1210 out tokens · 64458 ms · 2026-05-22T13:37:08.195555+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. SL(C)AMma: Simultaneous Localisation, (Calibration) and Mapping With a Magnetometer Array

    cs.RO 2026-04 unverdicted novelty 6.0

    Magnetometer-array SLAM with optional joint calibration delivers accurate indoor trajectories and over 80% drift reduction versus single-sensor or pure integration baselines on datasets where prior magnetic SLAM fails.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    Magnetic navigation using attitude-invariant magnetic field information for loop closure detection,

    N. Pavlasek, C. C. Cossette, D. Roy-Guay, and J. R. Forbes, “Magnetic navigation using attitude-invariant magnetic field information for loop closure detection,” in Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), Detroit, USA, Oct. 2023, pp. 5251–5257

    work page 2023

  2. [2]

    Magloc-ar: Magnetic-based localization for visual-free aug- mented reality in large-scale indoor environments,

    H. Liu, H. Xue, L. Zhao, D. Chen, Z. Peng, and G. Zhang, “Magloc-ar: Magnetic-based localization for visual-free aug- mented reality in large-scale indoor environments,” IEEE Trans. Vis. Comput. Graphics, vol. 29, no. 11, pp. 4383–4393, 2023

    work page 2023

  3. [3]

    MAINS: A magnetic-field-aided inertial navigation system for indoor positioning,

    C. Huang, G. Hendeby, H. Fourati, C. Prieur, and I. Skog, “MAINS: A magnetic-field-aided inertial navigation system for indoor positioning,” IEEE Sens. J., vol. 24, no. 9, pp. 15 156– 15 166, 2024

    work page 2024

  4. [4]

    Improving inertial veloc- ity estimation through magnetic field gradient-based extended Kalman filter,

    M. Zmitri, H. Fourati, and C. Prieur, “Improving inertial veloc- ity estimation through magnetic field gradient-based extended Kalman filter,” in Proc. Int. Conf. Indoor Positioning Indoor Navig. (IPIN), Pisa, Italy, Sep. 2019, pp. 1–7

    work page 2019

  5. [5]

    Enhancing visual inertial SLAM with magnetic measurements,

    B. Joshi and I. Rekleitis, “Enhancing visual inertial SLAM with magnetic measurements,” in Proc. IEEE Int. Conf. Robot. Autom. (ICRA), Yokohama, Japan, May 2024, pp. 10 012– 10 019

    work page 2024

  6. [6]

    Inertial Magnetic SLAM Systems Using Low-Cost Sensors

    C. Huang, G. Hendeby, and I. Skog, “Inertial magnetic Slam sys- tems using low-cost sensors,” arXiv preprint arXiv:2512.10128, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [7]

    SC-EKF-MIO: Magnetic-inertial odometry based on stochastic clone Extended Kalman filter,

    J. Li, J. Kuang, Y. Wang, and X. Niu, “SC-EKF-MIO: Magnetic-inertial odometry based on stochastic clone Extended Kalman filter,” IEEE Trans. Instrum. Meas., vol. 74, pp. 1–13, 2025

    work page 2025

  8. [8]

    An algorithm for the in-field calibration of a MEMS IMU,

    U. Qureshi and F. Golnaraghi, “An algorithm for the in-field calibration of a MEMS IMU,” IEEE Sensors Journal, vol. 17, no. 22, pp. 7479–7486, 2017

    work page 2017

  9. [9]

    Calibrating an ecompass in the presence of hard and soft-iron interference,

    T. Ozyagcilar, “Calibrating an ecompass in the presence of hard and soft-iron interference,” Freescale Semiconductor Ltd, pp. 1– 17, 2012

    work page 2012

  10. [10]

    Magnetometer calibration using inertial sensors,

    M. Kok and T. B. Schön, “Magnetometer calibration using inertial sensors,” IEEE Sens. J., vol. 16, no. 14, pp. 5679–5689, 2016

    work page 2016

  11. [11]

    An efficient method for gyroscope- aided full magnetometer calibration,

    M. Zhu, Y. Wu, and W. Yu, “An efficient method for gyroscope- aided full magnetometer calibration,” IEEE Sens. J., vol. 19, no. 15, pp. 6355–6361, 2019. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    work page 2019

  12. [12]

    Dynamic magnetometer calibration and alignment to inertial sensors by Kalman filter- ing,

    Y. Wu, D. Zou, P. Liu, and W. Yu, “Dynamic magnetometer calibration and alignment to inertial sensors by Kalman filter- ing,” IEEE Trans. Control Syst. Technol., vol. 26, no. 2, pp. 716–723, 2018

    work page 2018

  13. [13]

    Mag.i.c.al.–a unified method- ology for magnetic and inertial sensors calibration and align- ment,

    K. Papafotis and P. P. Sotiriadis, “Mag.i.c.al.–a unified method- ology for magnetic and inertial sensors calibration and align- ment,” IEEE Sens. J., vol. 19, no. 18, pp. 8241–8251, 2019

    work page 2019

  14. [14]

    Calibration of a magnetometer in combination with inertial sensors,

    M. Kok, J. D. Hol, T. B. Schön, F. Gustafsson, and H. Luinge, “Calibration of a magnetometer in combination with inertial sensors,” in Proc. Int. Conf. Inf. Fusion, Singapore, Jul. 2012, pp. 787–793

    work page 2012

  15. [15]

    Iterative calibration method for inertial and magnetic sensors,

    E. Dorveaux, D. Vissière, A.-P. Martin, and N. Petit, “Iterative calibration method for inertial and magnetic sensors,” in Proc. IEEE Conf. Decis. Control (CDC), Shanghai, China, Dec. 2009, pp. 8296–8303

    work page 2009

  16. [16]

    Särkkä, Bayesian Filtering and Smoothing, ser

    S. Särkkä, Bayesian Filtering and Smoothing, ser. Institute of Mathematical Statistics Textbooks. Cambridge University Press, 2013

    work page 2013

  17. [17]

    On- manifold preintegration for real-time visual–inertial odometry,

    C. Forster, L. Carlone, F. Dellaert, and D. Scaramuzza, “On- manifold preintegration for real-time visual–inertial odometry,” IEEE Trans. Robot., vol. 33, no. 1, pp. 1–21, 2016

    work page 2016

  18. [18]

    iSAM: Incremen- tal smoothing and mapping,

    M. Kaess, A. Ranganathan, and F. Dellaert, “iSAM: Incremen- tal smoothing and mapping,” IEEE Trans. Robot., vol. 24, no. 6, pp. 1365–1378, 2008

    work page 2008

  19. [19]

    Nonlinear filters: beyond the Kalman filter,

    F. Daum, “Nonlinear filters: beyond the Kalman filter,” IEEE Aerosp. Electron. Syst. Mag., vol. 20, no. 8, pp. 57–69, 2005

    work page 2005

  20. [20]

    G. S. Chirikjian, Stochastic models, information theory, and Lie groups, volume 2: Analytic methods and modern applications. Springer Science & Business Media, 2011, vol. 2

    work page 2011

  21. [21]

    Martin and H

    A. Martin and H. Anders, Optimization for Learning and Control. John Wiley & Sons, Ltd, 2023

    work page 2023

  22. [22]

    Inertial measurement unit error modeling tutorial: Inertial navigation system state estimation with real-time sensor calibration,

    J. A. Farrell, F. O. Silva, F. Rahman, and J. Wendel, “Inertial measurement unit error modeling tutorial: Inertial navigation system state estimation with real-time sensor calibration,” IEEE Control Syst. Mag., vol. 42, no. 6, pp. 40–66, 2022

    work page 2022

  23. [23]

    Zero- velocity detection–an algorithm evaluation,

    I. Skog, P. Handel, J.-O. Nilsson, and J. Rantakokko, “Zero- velocity detection–an algorithm evaluation,” IEEE Trans. Biomed. Eng., vol. 57, no. 11, pp. 2657–2666, 2010

    work page 2010

  24. [24]

    Geomagnetic calculator,

    NOAA National Centers for Environmental Information, “Geomagnetic calculator,” 2025, accessed: 31-Mar-2025. [Online]. A vailable: https://www.ngdc.noaa.gov/geomag/ calculators/magcalc.shtml#igrfwmm

    work page 2025

  25. [25]

    On misalignment between magnetometer and inertial sensors,

    Y. Wu and S. Luo, “On misalignment between magnetometer and inertial sensors,” IEEE Sens. J., vol. 16, no. 16, pp. 6288– 6297, 2016

    work page 2016

  26. [26]

    Ljung, System identification (2nd ed.): theory for the user

    L. Ljung, System identification (2nd ed.): theory for the user. USA: Prentice Hall PTR, 1999

    work page 1999

  27. [27]

    Time synchronization errors in loosely coupled gps-aided inertial navigation systems,

    I. Skog and P. Handel, “Time synchronization errors in loosely coupled gps-aided inertial navigation systems,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, pp. 1014–1023, 2011

    work page 2011

  28. [28]

    C. M. Bishop and N. M. Nasrabadi, Pattern recognition and machine learning. Springer, 2006, vol. 4, no. 4

    work page 2006

  29. [29]

    Fitting models to data: Residual analysis, a primer,

    J. Martin, D. D. R. De Adana, and A. G. Asuero, “Fitting models to data: Residual analysis, a primer,” Uncertainty quantification and model calibration, vol. 133, 2017

    work page 2017

  30. [30]

    Manopt, a Matlab toolbox for optimization on manifolds,

    N. Boumal, B. Mishra, P.-A. Absil, and R. Sepulchre, “Manopt, a Matlab toolbox for optimization on manifolds,” J. Mach. Learn. Res., vol. 15, no. 42, pp. 1455–1459, 2014. [Online]. A vailable:https://www.manopt.org

    work page 2014

  31. [31]

    A micro lie theory for state estimation in robotics,

    J. Sola, J. Deray, and D. Atchuthan, “A micro lie theory for state estimation in robotics,” arXiv preprint arXiv:1812.01537, 2018. Chuan Huang (Student member, IEEE) re- ceived the B.Sc. from Beihang University in 2018 and the M.Sc. degree from China Electronics Technology Group Corporation Academy of Electronics and Information Tech- nology in 2021. From...

    work page arXiv 2018