pith. sign in

arxiv: 1906.10783 · v1 · pith:EDJCUIIKnew · submitted 2019-06-25 · 💻 cs.RO

OLAE-ICP: Robust and fast alignment of geometric features with the optimal linear attitude estimator

Jose Luis Blanco-Claraco This is my paper

Pith reviewed 2026-05-25 16:08 UTC · model grok-4.3

classification 💻 cs.RO
keywords point cloud registrationWahba problemgeometric feature alignmentICPattitude estimationSE(3) transformationOLAE
0
0 comments X

The pith

Any point-cloud or Wahba solver can align mixed sets of points, lines and planes by reducing them all to vector observations.

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

The paper shows how to convert observations of points, lines, and planes into a single set of vector observations. This reduction lets any existing solver for point-cloud registration or Wahba's attitude problem compute the rigid SE(3) transform that minimizes the alignment cost between two such sets. The same framework supports efficient linear methods such as OLAE, outlier rejection via scale mismatch, and optional per-feature weights. Experiments confirm the approach works across noise levels and serves as the core of an ICP-style registration algorithm on real data.

Core claim

Any solution to point-cloud registration or Wahba's problem can be used to find the SE(3) transformation between two sets of geometric feature observations (points, lines, planes) that minimizes the corresponding cost function.

What carries the argument

Reduction of heterogeneous geometric primitives (points, lines, planes) to a common vector-observation framework for attitude estimation.

If this is right

  • The optimal attitude can be recovered with Horn's quaternion method, OLAE solving a small linear system, or a Gauss-Newton iteration.
  • Gross outliers in correspondences are rejected by detecting inconsistent scale in the estimated transformation.
  • Per-primitive relative weights and an optional robust loss can be applied when an initial guess is available.
  • The unified framework serves as the core of an iterative closest point or primitive algorithm that registers real datasets.

Where Pith is reading between the lines

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

  • The same reduction step would allow any future improvement to a Wahba solver to be applied immediately to mixed-feature registration without new derivations.
  • Sequential rotations to avoid Gibbs-vector singularities can be reused across any application that employs OLAE on geometric features.

Load-bearing premise

Converting lines and planes into vector observations preserves the optimality properties of the underlying attitude estimator without introducing systematic biases from feature-type differences.

What would settle it

Run the method on synthetic data where the true optimal alignment is known by direct minimization of point-to-line and point-to-plane distances, then check whether the recovered transform achieves the same minimum cost.

Figures

Figures reproduced from arXiv: 1906.10783 by Jose Luis Blanco-Claraco.

Figure 1
Figure 1. Figure 1: Errors for the optimal transformation estimated by three different methods (OLAE, [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Errors for the optimal transformation estimated by three different methods (OLAE, [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Errors for the optimal transformation estimated by three different methods (OLAE, [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Optimal transformation error statistics with and without the scale-based absolute outlier [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Error for different outlier ratios with and without the robust loss function to modify pair [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Results for the entire ICP system, when using each of the different algorithms as optimal [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
read the original abstract

The problems of point-cloud registration and attitude estimation from vector observations (Wahba's problem) have widespread applications in computer vision and mobile robotics. This work introduces a simple approach for integrating sets of geometric feature observations (points, lines, and planes) in such a way that any solution to either point-cloud registration or to Wahba's problem can be used to find the SE(3) transformation between the two sets that minimizes the corresponding cost function. We compare the performance of three solutions: classic Horn's optimal quaternion method, Optimal Linear Attitude Estimator (OLAE) that efficiently recovers the optimal Gibbs-Rodrigues vector solving a small linear system, and an iterative non-linear Gauss-Newton solver. Special care is given to explain how to overcome the Gibbs vector singularity for OLAE by using the method of sequential rotations. Gross outliers in point-to-point correspondences can be discarded by means of detecting transformation scale mismatches. The approach also allows the introduction of per-primitive relative weights, including an optional robust loss function that is applicable only if an initial guess for the solution is known in advance. Experiments are presented to evaluate how the three solutions tolerate noise in the input data for different kinds of geometric primitives. Finally, experiments with real datasets validate the suitability of the optimal alignment algorithm as the core of an Iterative Closest Point/Primitive (ICP) algorithm. An open-source implementation of all the described algorithms is provided in https://github.com/MOLAorg/mp2p_icp

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 claims that geometric features (points, lines, planes) can be mapped to vector observations such that any solver for point-cloud registration or Wahba's problem (Horn's quaternion method, OLAE via linear system on Gibbs vector, or Gauss-Newton) recovers the SE(3) transformation minimizing the corresponding feature cost; it adds outlier rejection via scale mismatch, per-primitive weights, and an optional robust loss, with experiments on synthetic noise and real ICP datasets plus open-source code.

Significance. If the mapping preserves exact optimality for mixed feature types, the work unifies established attitude estimators with heterogeneous registration, enabling efficient non-iterative or lightly iterative solutions in robotics and vision pipelines. The open-source implementation at the cited GitHub repository is a clear strength for reproducibility. The result would be of moderate significance as an engineering bridge rather than a new theoretical primitive, provided the equivalence is rigorously shown.

major comments (2)
  1. [Approach for integrating geometric feature observations (abstract and §3)] The central claim (abstract and method description) is that the vector-observation mapping for lines (direction vectors) and planes (normals) yields a cost whose joint minimum over rotation+translation coincides with the original geometric feature cost when features are heterogeneous and per-primitive weights are used. No derivation, cost-function equivalence proof, or counter-example analysis is supplied to confirm that the attitude cost on mapped vectors plus separate positional terms recovers the intended minimizer; this is load-bearing for the assertion that 'any solution to point-cloud registration or Wahba's problem can be used'.
  2. [OLAE integration and outlier handling] § on OLAE with sequential rotations and the scale-mismatch outlier rule: the manuscript must demonstrate that the combined weighted cost after mapping has its minimum at the same (R,t) recovered by the vector solver; otherwise the claim that the method 'minimizes the corresponding cost function' for mixed primitives does not hold.
minor comments (2)
  1. [Abstract] The abstract states that experiments evaluate tolerance to noise 'for different kinds of geometric primitives' but does not report the specific noise models or quantitative metrics (e.g., rotation/translation error tables) used; adding these would improve clarity.
  2. [Weights and robust loss description] Notation for per-primitive weights and the robust loss should be introduced with an explicit equation when first mentioned, to avoid ambiguity when the loss is said to be 'applicable only if an initial guess is known'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on the central claims of the paper. We address each major comment below and will revise the manuscript accordingly to include the requested derivations and demonstrations of cost equivalence.

read point-by-point responses

  1. Referee: [Approach for integrating geometric feature observations (abstract and §3)] The central claim (abstract and method description) is that the vector-observation mapping for lines (direction vectors) and planes (normals) yields a cost whose joint minimum over rotation+translation coincides with the original geometric feature cost when features are heterogeneous and per-primitive weights are used. No derivation, cost-function equivalence proof, or counter-example analysis is supplied to confirm that the attitude cost on mapped vectors plus separate positional terms recovers the intended minimizer; this is load-bearing for the assertion that 'any solution to point-cloud registration or Wahba's problem can be used'.

    Authors: We agree that the manuscript would be strengthened by an explicit derivation of the cost equivalence. The mappings are constructed so that the rotation-dependent portion of each geometric feature cost (point-to-point, line direction, plane normal) reduces to a vector observation error in the Wahba formulation, with translation recovered separately via the centroid or similar. In the revision we will add a dedicated subsection in §3 deriving the equivalence for homogeneous and heterogeneous cases, including per-primitive weights, and briefly discuss why the joint (R,t) minimizer is recovered. A short counter-example check for a mixed point-line case will also be included. revision: yes

  2. Referee: [OLAE integration and outlier handling] § on OLAE with sequential rotations and the scale-mismatch outlier rule: the manuscript must demonstrate that the combined weighted cost after mapping has its minimum at the same (R,t) recovered by the vector solver; otherwise the claim that the method 'minimizes the corresponding cost function' for mixed primitives does not hold.

    Authors: Because OLAE (and the other solvers) exactly minimize the mapped Wahba cost, the equivalence derivation will directly imply that the recovered (R,t) also minimizes the original feature cost. Sequential rotations preserve optimality by selecting the representation that avoids the singularity while yielding the same rotation matrix. The scale-mismatch rule is a post-processing filter on correspondences and does not alter the minimizer of the retained set. In the revision we will add an explicit statement and short proof in the OLAE subsection confirming that the minimum of the combined weighted mapped cost coincides with the original geometric cost, covering the outlier-rejection case as well. revision: yes

Circularity Check

0 steps flagged

No circularity: mapping reduces new problem to external established solvers without self-referential reduction.

full rationale

The paper's derivation introduces a mapping of points/lines/planes to vector observations so that any existing point-cloud registration or Wahba solver (Horn, OLAE, Gauss-Newton) can be applied directly. This is a reduction to independent external methods whose optimality properties are taken as given, not redefined or fitted within the paper. No equations or claims reduce a result to its own inputs by construction, no parameters are fitted to the target outcome and renamed as predictions, and no load-bearing uniqueness theorem is imported via self-citation. The approach is self-contained against the cited external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions from attitude estimation and registration literature; no new free parameters, ad-hoc axioms, or invented entities are introduced in the abstract.

axioms (1)
  • domain assumption Geometric features (points, lines, planes) can be represented as vector observations suitable for Wahba's problem without loss of optimality.
    Invoked when stating that any Wahba solver can be applied after integration.

pith-pipeline@v0.9.0 · 5799 in / 1260 out tokens · 38082 ms · 2026-05-25T16:08:55.387259+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

15 extracted references · 15 canonical work pages · 1 internal anchor

  1. [1]

    Method for registration of 3-d shapes

    [Besl and McKay(1992)] Paul J Besl and Neil D McKay. Method for registration of 3-d shapes. In Sensor Fusion IV: Control Paradigms and Data Structures , volume 1611, pages 586–607. International Society for Optics and Photonics,

    work page 1992

  2. [2]

    A review of point cloud registration algorithms for mobile robotics

    [Pomerleau et al.(2015)Pomerleau, Colas, Siegwart, et al.] Fran¸ cois Pomerleau, Francis Co- las, Roland Siegwart, et al. A review of point cloud registration algorithms for mobile robotics. Foundations and Trends R⃝ in Robotics, 4(1):1–104,

    work page 2015

  3. [3]

    Generalized-icp

    [Segal et al.(2009)Segal, Haehnel, and Thrun] Aleksandr Segal, Dirk Haehnel, and Sebastian Thrun. Generalized-icp. In Robotics: science and systems , volume 2, page 435,

    work page 2009

  4. [4]

    The representation, recognition, and locating of 3-d objects

    [Faugeras and Hebert(1986)] Olivier D Faugeras and Martial Hebert. The representation, recognition, and locating of 3-d objects. The international journal of robotics research , 5(3):27–52,

    work page 1986

  5. [5]

    Closed-form solution of absolute orientation using unit quaternions

    11 [Horn(1987)] Berthold KP Horn. Closed-form solution of absolute orientation using unit quaternions. Josa a , 4(4):629–642,

    work page 1987

  6. [6]

    A least squares estimate of spacecraft attitude

    [Whaba(1965)] G Whaba. A least squares estimate of spacecraft attitude. SIAM Review , 7 (3):409,

    work page 1965

  7. [7]

    How to estimate at- titude from vector observations

    [Markley and Mortari(1999)] F Landis Markley and Daniele Mortari. How to estimate at- titude from vector observations. In AAS/AIAA Astrodynamics Specialist Conference ,

    work page 1999

  8. [8]

    Quaternion attitude estimation using vector observations

    [Markley and Mortari(2000)] F Landis Markley and Daniele Mortari. Quaternion attitude estimation using vector observations. Journal of the Astronautical Sciences , 48(2):359– 380,

    work page 2000

  9. [9]

    Optimal linear attitude estimator

    [Mortari et al.(2007)Mortari, Markley, and Singla] Daniele Mortari, F Landis Markley, and Puneet Singla. Optimal linear attitude estimator. Journal of Guidance, Control, and Dynamics, 30(6):1619–1627,

    work page 2007

  10. [10]

    The vectorial pa- rameterization of rotation

    [Bauchau and Trainelli(2003)] Olivier A Bauchau and Lorenzo Trainelli. The vectorial pa- rameterization of rotation. Nonlinear dynamics, 32(1):71–92,

    work page 2003

  11. [11]

    Efficient absolute ori- entation revisited

    [Lourakis and Terzakis(2018)] Manolis Lourakis and George Terzakis. Efficient absolute ori- entation revisited. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 5813–5818. IEEE,

    work page 2018

  12. [12]

    [Shuster and Oh(1981)] M.D

    URL https://ieeexplore.ieee.org/ abstract/document/8594296/. [Shuster and Oh(1981)] M.D. Shuster and S.D. Oh. Three-axis attitude determination from vector observations. Journal of Guidance, Control, and Dynamics , 4(1):70–77,

    work page arXiv 1981

  13. [13]

    A tutorial on se(3) transformation parameterizations and on-manifold optimization

    [Blanco(2010)] Jose-Luis Blanco. A tutorial on se(3) transformation parameterizations and on-manifold optimization. University of Malaga, Tech. Rep , 3,

    work page 2010

  14. [14]

    A volumetric method for building complex models from range images

    [Curless and Levoy(1996)] Brian Curless and Marc Levoy. A volumetric method for building complex models from range images

    work page 1996

  15. [15]

    A Polynomial-time Solution for Robust Registration with Extreme Outlier Rates

    [Yang and Carlone(2019)] Heng Yang and Luca Carlone. A polynomial-time solution for robust registration with extreme outlier rates. arXiv preprint arXiv:1903.08588 ,

    work page internal anchor Pith review Pith/arXiv arXiv 2019