pith. sign in

arxiv: 2604.04792 · v1 · submitted 2026-04-06 · 📡 eess.SP

Multi-Scaled Unscented Kalman Filter

Amit Levy , Itzik Klein This is my paper

Pith reviewed 2026-05-10 18:49 UTC · model grok-4.3

classification 📡 eess.SP
keywords unscented Kalman filtermulti-scaled UKFsigma pointsnonlinear state estimationdynamic systemsscaled unscented transformfiltering algorithmsmulti-dimensional models
0
0 comments X

The pith

The multi-scaled unscented Kalman filter allows independent scaling of sigma points for each state dimension.

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

This paper shows how to modify the unscented Kalman filter so that each dimension of a multi-dimensional state can have its own scaling parameters when generating sigma points. Standard UKF forces every dimension to share the same scaling values, which restricts how well the filter can match systems where different states evolve at different rates or scales. The authors supply a mathematical construction that keeps the sigma points symmetric and correctly weighted while letting the spreads vary. They test the change on two nonlinear dynamic systems and report tighter estimates of the posterior mean and covariance. Anyone who tunes UKF for navigation, control, or sensor fusion problems would see the practical value in removing that shared-parameter constraint.

Core claim

The multi-scaled UKF enables spreading differently per state, while maintaining the key properties of the sigma points and UKF. A rigorous mathematical foundation is provided, introducing a novel theoretical approach to multi-scaling. The benefits of this approach are demonstrated through two distinct nonlinear dynamic systems. Consequently, our multi-scaled UKF captures the nonlinear behavior of multi-dimensional states more effectively, leading to improved estimation accuracy.

What carries the argument

The multi-scaling mechanism added to the scaled unscented transform, which replaces the single shared set of scaling parameters with a per-dimension vector of scaling values while preserving sigma-point symmetry and weighting.

If this is right

  • The filter can match the spread of sigma points to the actual scale of each state without manual compromise across dimensions.
  • Estimation accuracy improves in systems whose state components evolve on different time or amplitude scales.
  • The sigma-point mean and covariance calculations remain algebraically identical in form to the original UKF.
  • The method extends directly to any application already using UKF for multi-state nonlinear filtering.

Where Pith is reading between the lines

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

  • The same per-dimension scaling idea could be inserted into other sigma-point or quadrature filters without changing their core sampling logic.
  • In high-dimensional problems the approach might reduce the amount of hand-tuning needed when state components have mismatched dynamics.
  • An online version that adapts the per-state scales from incoming measurements could be developed as a direct extension.

Load-bearing premise

States in multi-dimensional models often exhibit substantially different behaviors that cannot be captured well by one common set of scaling parameters.

What would settle it

Apply both the standard UKF and the multi-scaled UKF to the same multi-dimensional nonlinear test system and measure whether the multi-scaled version produces larger or equal root-mean-square estimation errors.

Figures

Figures reproduced from arXiv: 2604.04792 by Amit Levy, Itzik Klein.

Figure 1
Figure 1. Figure 1: The MS-UKF flowchart. The new MS parameters [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sigma points extraction for x1 as function of α (0.01 vs 1.0) - larger α spreads fits the model better [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sigma points extraction for x2 as function of α (0.01 vs 1.0) - smaller α spreads fits the model better [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: TSTD comparison between two UKFs with different [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sigma points estimation due to α = 0.01 and α = 1.0 after one cycle of example 2 x1 axis [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: TSTD comparison between a UKF with a single [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

The unscented Kalman filter (UKF) is a commonly used algorithm capable of estimating the states of nonlinear dynamic systems. It carefully chooses a set of sample points, called sigma points that capture the nonlinear system states posterior mean and covariance. The filter is based on the scaled unscented transform, where the scaling parameters impact the spreading of the sigma points, determining the estimated model capturing. In its current form, the UKF employs a single set of scaling parameters shared by all sigma points. Because states in multi-dimensional models often exhibit substantially different behaviors, this imposes a critical limitation: the standard UKF parameters cannot be tuned to extend the spread for one dimension while reducing it for another. To bridge this gap, we propose the multi-scaled UKF to enable spreading differently per state, while maintaining the key properties of the sigma points and UKF. A rigorous mathematical foundation is provided, introducing a novel theoretical approach to multi-scaling. The benefits of this approach are demonstrated through two distinct nonlinear dynamic systems. Consequently, our multi-scaled UKF captures the nonlinear behavior of multi-dimensional states more effectively, leading to improved estimation accuracy.

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 manuscript proposes a multi-scaled Unscented Kalman Filter (UKF) that replaces the standard single scalar scaling parameter with per-state scaling parameters. This is intended to allow independent control of sigma-point spread in each dimension of a multi-dimensional nonlinear system. The paper supplies a mathematical derivation for the multi-scaled sigma points and weights, asserts that the defining mean- and covariance-matching properties of the unscented transform are preserved, and reports improved estimation accuracy on two example nonlinear dynamic systems.

Significance. If the multi-scaled construction rigorously preserves the weighted-mean and weighted-covariance identities of the unscented transform, the method would remove a practical limitation of the classical UKF for heterogeneous state dynamics and could be adopted in tracking, navigation, and control applications. The explicit claim of a parameter-free or property-preserving derivation is therefore the load-bearing element of the contribution.

major comments (2)
  1. [theoretical approach / multi-scaling derivation] The central claim that per-dimension scaling preserves the UKF sigma-point identities must be shown explicitly. In the section presenting the novel theoretical approach to multi-scaling, the derivation should recompute the weighted sum of the sigma points and the weighted outer-product sum after the vector-valued scaling is introduced, and demonstrate that cross terms cancel for correlated states. The abstract asserts that the key properties are maintained, but the provided skeptic analysis indicates this step is not yet secured.
  2. [multi-scaled sigma-point construction] Because the standard UKF weights are derived from a single scalar lambda that appears uniformly, replacing lambda by a vector changes the point locations asymmetrically. The manuscript must state whether the weights themselves are redefined dimension-wise or left unchanged; if unchanged, the covariance-matching identity fails when off-diagonal covariances are nonzero. This point is load-bearing for the accuracy-improvement claim.
minor comments (2)
  1. Notation for the new per-state scaling vector should be introduced with a clear distinction from the classical scalar parameters (alpha, beta, kappa) to prevent reader confusion.
  2. The two demonstration examples should report both the classical UKF and the multi-scaled UKF with identical tuning effort so that the accuracy gain can be attributed to the multi-scaling rather than to extra degrees of freedom.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and valuable comments on our manuscript. We have carefully considered the points raised regarding the theoretical derivation of the multi-scaled UKF and will revise the manuscript to address them explicitly. Our responses to the major comments are as follows.

read point-by-point responses

  1. Referee: [theoretical approach / multi-scaling derivation] The central claim that per-dimension scaling preserves the UKF sigma-point identities must be shown explicitly. In the section presenting the novel theoretical approach to multi-scaling, the derivation should recompute the weighted sum of the sigma points and the weighted outer-product sum after the vector-valued scaling is introduced, and demonstrate that cross terms cancel for correlated states. The abstract asserts that the key properties are maintained, but the provided skeptic analysis indicates this step is not yet secured.

    Authors: We appreciate this suggestion for strengthening the presentation. Our derivation in the manuscript constructs the multi-scaled sigma points by applying a diagonal scaling matrix to the standard deviations, and the weights are computed based on the original formulation. To make the preservation explicit, we will revise the section to include direct recomputation of the weighted mean (which remains the state mean by symmetry) and the weighted covariance sum. We will show that for the outer-product terms, the contributions from positive and negative sigma points cancel the cross terms appropriately, preserving the covariance even when states are correlated. This will be added as a new lemma or proposition with the full algebraic expansion. revision: yes

  2. Referee: [multi-scaled sigma-point construction] Because the standard UKF weights are derived from a single scalar lambda that appears uniformly, replacing lambda by a vector changes the point locations asymmetrically. The manuscript must state whether the weights themselves are redefined dimension-wise or left unchanged; if unchanged, the covariance-matching identity fails when off-diagonal covariances are nonzero. This point is load-bearing for the accuracy-improvement claim.

    Authors: We agree that the manuscript should explicitly address the treatment of the weights. In the proposed multi-scaled UKF, the weights are left unchanged from the standard UKF (derived using the scalar lambda), while the sigma point locations are adjusted using per-state scaling factors. We will add a clear statement to this effect in the revised manuscript. Furthermore, we will provide the explicit verification that the covariance-matching property holds for nonzero off-diagonal elements by computing the weighted sum of outer products, demonstrating that the scaling is applied in a manner consistent with the original covariance structure. This addresses the concern and supports the accuracy improvements observed in the examples. revision: yes

Circularity Check

0 steps flagged

No significant circularity; direct extension of scaled UKF with independent math foundation

full rationale

The paper defines a per-dimension scaling vector for sigma points in the unscented transform and asserts that the weighted mean and covariance properties are preserved under this change. This is presented as a novel theoretical construction with explicit equations for the multi-scaled points, not as a fit to data or a renaming of prior results. No load-bearing self-citations, no fitted parameters renamed as predictions, and no self-definitional loops appear in the derivation chain. The accuracy claims rest on numerical demonstrations on two example systems rather than on tautological re-derivation of the inputs. The skeptic concern about covariance matching is a question of proof correctness, not a reduction of the claimed result to its own inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that multi-scaling can be formulated to preserve the mean and covariance capture properties of the unscented transform, with per-state scaling parameters serving as user-chosen design choices rather than data-fitted values.

free parameters (1)
  • per-state scaling parameters
    Scaling parameters (analogous to alpha, beta, kappa but allowed to differ per dimension) are design choices selected by the user to control sigma point spread for each state.
axioms (1)
  • domain assumption Multi-scaling maintains the key statistical properties of the sigma points and UKF
    The paper states that the approach preserves these properties while enabling different spreads per dimension.

pith-pipeline@v0.9.0 · 5488 in / 1136 out tokens · 34551 ms · 2026-05-10T18:49:45.658585+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

  1. [1]

    A New Approach to Linear Filtering and Prediction Problems,

    R. E. Kalman, “A New Approach to Linear Filtering and Prediction Problems,”Journal of Basic Engineering, vol. 82, no. 1, pp. 35–45, 1960

    work page 1960

  2. [2]

    Bar-Shalom, X

    Y . Bar-Shalom, X. R. Li, and T. Kirubarajan,Estimation with applications to tracking and navigation: Theory algorithms and software.John Wiley & Sons, 2004

    work page 2004

  3. [3]

    The unscented Kalman filter for nonlinear estimation,

    E. A. Wan and R. V . D. Merwe, “The unscented Kalman filter for nonlinear estimation,”Proceedings of the IEEE 2000 Adaptive Systems for Signal Processing, Commu- nications, and Control Symposium, pp. 153–158, 2000

    work page 2000

  4. [4]

    A New Extension of the Kalman Filter to Nonlinear Systems,

    S. J. Julier and J. K. Uhlmann, “A New Extension of the Kalman Filter to Nonlinear Systems,”Signal Processing, Sensor Fusion, and Target Recognition VI, vol. Proc. SPIE 3068, 1997

    work page 1997

  5. [5]

    Unscented filtering and non- linear estimation,

    S. Julier and J. Uhlmann, “Unscented filtering and non- linear estimation,”Proceedings of the IEEE, vol. 92, no. 3, pp. 401–422, 2004

    work page 2004

  6. [6]

    Applying the unscented Kalman filter for nonlinear state estimation,

    R. Kandepu, B. Foss, and L. Imsland, “Applying the unscented Kalman filter for nonlinear state estimation,” Journal of process control, vol. 18, no. 7-8, pp. 753–768, 2008

    work page 2008

  7. [7]

    An unscented Kalman filter for in-motion alignment of low-cost imus,

    E.-H. Shin and N. El-Sheimy, “An unscented Kalman filter for in-motion alignment of low-cost imus,” in PLANS 2004. Position Location and Navigation Sympo- sium (IEEE Cat. No. 04CH37556). IEEE, 2004, pp. 273–279

    work page 2004

  8. [8]

    Comparison of unscented Kalman filters,

    Y . Hao, Z. Xiong, F. Sun, and X. Wang, “Comparison of unscented Kalman filters,” in2007 international confer- ence on mechatronics and automation. IEEE, 2007, pp. 895–899

    work page 2007

  9. [9]

    Modified unscented Kalman filtering and its application in autonomous satellite navigation,

    K. Xiong, L. Liu, and H. Zhang, “Modified unscented Kalman filtering and its application in autonomous satellite navigation,”Aerospace Science and Technology, vol. 13, no. 4-5, pp. 238–246, 2009

    work page 2009

  10. [10]

    Adaptive Neural Unscented Kalman Filter,

    A. Levy and I. Klein, “Adaptive Neural Unscented Kalman Filter,”IEEE Transactions on Intelligent V ehi- cles, vol. 11, no. 3, pp. 1–10, 2026

    work page 2026

  11. [11]

    A Hybrid Neural-Assisted Unscented Kalman Filter for Unmanned Ground Vehicle Navigation,

    G. Versano and I. Klein, “A Hybrid Neural-Assisted Unscented Kalman Filter for Unmanned Ground Vehicle Navigation,”arXiv preprint arXiv:2603.11649, 2026

    work page arXiv 2026

  12. [12]

    Unscented Kalman Filter with a Nonlinear Propagation Model for Navigation Ap- plications,

    A. Levy and I. Klein, “Unscented Kalman Filter with a Nonlinear Propagation Model for Navigation Ap- plications,” in2025 IEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea), 2025, pp. 509–514. 11

    work page 2025

  13. [13]

    Optimized selection of sigma points in the unscented Kalman filter,

    Y . Cheng and Z. Liu, “Optimized selection of sigma points in the unscented Kalman filter,”International Con- ference on Electrical and Control Engineering, Yichang, China, pp. 3073–3075, 2011

    work page 2011

  14. [14]

    A Systematization of the Unscented Kalman Filter Theory,

    H. M. T. Menegaz, J. Y . Ishihara, G. A. Borges, and A. N. Vargas, “A Systematization of the Unscented Kalman Filter Theory,”IEEE Transactions on Automatic Control, pp. 2583–2598, 2015

    work page 2015

  15. [15]

    Modification of unscented Kalman filter using a set of scaling param- eters,

    S. M.-B. M. Mahboubeh Zarei-Jalalabadi, “Modification of unscented Kalman filter using a set of scaling param- eters,”IET Signal Process), 2018

    work page 2018

  16. [16]

    The scaled unscented transformation,

    S. J. Julier, “The scaled unscented transformation,”Pro- ceedings of the 2002 American Control Conference, vol. 6, pp. 4555–4559, 2002

    work page 2002

  17. [17]

    Scaling parameter in unscented transform: Analysis and specification,

    O. Straka, J. Dunik, and M. Simandl, “Scaling parameter in unscented transform: Analysis and specification,”2012 American Control Conference (ACC), Montreal, QC, Canada, vol. 6, pp. 5550–5555, 2012

    work page 2012

  18. [18]

    Unscented Kalman filter with advanced adaptation of scaling parameter,

    O. Straka, J. Dun ´ık, and M.ˇSimandl, “Unscented Kalman filter with advanced adaptation of scaling parameter,” Automatica, vol. 50, no. 10, pp. 2657–2664, 2014

    work page 2014

  19. [19]

    Scaling parameters selection principle for the scaled unscented Kalman filter,

    N. Yongfang and Z. Tao, “Scaling parameters selection principle for the scaled unscented Kalman filter,”IEEE Intelligent V ehicles Symposium, vol. 29, no. 3, pp. 601– 610, 2018

    work page 2018

  20. [20]

    Dynamic causal modeling,

    K. Friston, L. Harrison, and W. Penny, “Dynamic causal modeling,”NeuroImage, vol. 19, no. 4, pp. 1273–1302, 2003

    work page 2003