Higher-Order Cayley Transforms with Applications to Attitude Representations

Panagiotis Tsiotras, John L. Junkins, Hanspeter Schaub

TL;DR

This paper generalizes the classical Cayley transform, a fundamental tool for representing rotations, to higher orders. By drawing an analogy between complex numbers and matrices, the authors introduce "higher-order Cayley transforms" that offer new ways to parameterize rotations, potentially leading to more robust and versatile attitude representations in fields like aerospace engineering.

What Problem Does This Paper Address?

Representing the orientation of a rigid body (like a spacecraft or robot arm) in 3D space is a fundamental problem in many engineering disciplines. This is typically done using rotation matrices, which belong to the special orthogonal group SO(3). While these matrices are accurate, they are redundant (9 numbers to describe 3 degrees of freedom). To simplify calculations and analysis, various parameterizations have been developed, such as Euler angles and Rodrigues parameters.

The classical Cayley transform provides a way to map skew-symmetric matrices (which have 3 independent parameters) to proper orthogonal matrices (rotations). However, this transform has limitations, particularly when dealing with certain rotations (specifically, those involving a rotation of 180 degrees about an axis). This paper addresses the need for more comprehensive parameterizations that can overcome these limitations and potentially offer advantages in terms of numerical stability and range of applicability. The authors are motivated by recent work on "modified Rodrigues parameters," which showed that a second-order Cayley transform could generate these improved parameters.

Methods

The core of the paper lies in generalizing the Cayley transform using the concept of conformal mappings from complex analysis. The authors establish an analogy:

  • Complex Numbers: Real numbers (IR), imaginary numbers (iIR), and the unit circle (S¹).
  • Matrices: Symmetric matrices, skew-symmetric matrices, and proper orthogonal matrices (SO(n)).

They then extend the classical Cayley transform, which is a bilinear transformation in complex numbers, to "higher-order Cayley transforms" in the matrix space. This is achieved by composing the standard Cayley transform with a higher-order mapping (like $z^n$) in the complex plane and then applying this to matrices. Specifically, they define higher-order Cayley transforms as:

$C = (I - Q)^n (I + Q)^{-n}$

where $Q$ is a skew-symmetric matrix and $n$ is the order of the transform. This generalization allows them to derive new sets of parameters, which they call "higher-order Rodrigues parameters," by inverting these transforms. They explicitly derive the forms for third and fourth-order transforms and their corresponding parameters.

Datasets

Not discussed in the paper. The paper focuses on theoretical derivations and parameterizations, with numerical examples presented in Section 7 to illustrate the behavior of the derived parameters.

Key Insights and Ideas

  • Matrix-Complex Number Analogy: The paper elegantly formalizes the intuitive connection between real/imaginary numbers and symmetric/skew-symmetric matrices, and between the unit circle and orthogonal matrices. This analogy is key to generalizing complex analytic concepts to matrix spaces.
  • Conformal Mappings as the Unifying Principle: The authors show that the classical Cayley transform is a specific instance of a conformal mapping. By generalizing these conformal mappings to higher orders in the matrix domain, they can systematically generate new attitude parameterizations.
  • Higher-Order Transforms for Extended Range: The primary motivation and outcome is the generation of "higher-order Rodrigues parameters." These new parameters are shown to extend the range of valid rotations that can be represented without encountering singularities, a significant improvement over classical methods.
  • "Shadow" Parameters: The paper introduces the concept of "shadow" parameter sets, which are alternative representations for the same physical orientation. This duality is particularly useful for maintaining continuity and avoiding singularities, especially in higher-order parameterizations.
  • Linearization of Kinematics: The higher-order parameters, particularly the fourth-order ones, exhibit a more "linear-like" behavior with respect to the principal rotation angle, simplifying kinematic analysis and potentially improving numerical integration.

Key Results

  • Generalization of Cayley Transform: The paper formally defines higher-order Cayley transforms of the form $C = (I - Q)^n (I + Q)^{-n}$.
  • Derivation of Higher-Order Rodrigues Parameters: Explicit formulas for third and fourth-order Rodrigues-like parameters (denoted as $p$ and $\tau$, respectively) are derived.
  • Improved Singularity Avoidance: Numerical simulations (Figure 3 and 4) demonstrate that the higher-order parameters ($\tau$ and $p$) exhibit better behavior and avoid singularities for larger rotation angles compared to classical Rodrigues parameters ($\rho$) and Modified Rodrigues parameters ($\sigma$). The "shadow" parameter sets further enhance this capability.
  • More Linear Kinematics: The kinematic equations for the higher-order parameters are shown to be well-defined and exhibit a more linear behavior, which is beneficial for attitude dynamics and control.

Limitations and Future Work

  • Applicability of Higher-Order Parameters: The authors acknowledge that while they have generated these higher-order parameters, their practical applicability in real-world attitude problems still needs to be determined.
  • Kinematics of Higher-Order Parameters: While the kinematics are discussed and derived, a more in-depth investigation of their behavior and implications is left for future work.
  • Numerical Stability: Although the higher-order parameters show improved singularity avoidance, the paper notes that finite escape times can still occur due to nonlinearities in the kinematic equations, similar to classical parameterizations.
  • Extension to Higher Orders: The paper suggests that the approach can be extended to any order $2k$ Cayley transform, implying a vast space of potential parameterizations to explore.

Summary updated

Abstract

We generalize previous results on attitude representations by using Cayley transforms. First, we show that proper orthogonalmatrices, which naturally represent rotations, can be generated by a form of conformal analytic mappings in the space of matrices. Using a natural parallelism between the elements of the complex plane and the realmatrices, we generate higher-orderCayley transformsanddiscuss someof their properties. These higher-order Cayley transformsare shownto parameterizeproper orthogonalmatrices intohigher-orderRodriguesparameters.
BibTeX
@article{tsiotras1997higher,
  author = {Panagiotis Tsiotras and John L. Junkins and Hanspeter Schaub},
  title = {{Higher-Order Cayley Transforms with Applications to Attitude Representations}},
  journal = {Journal of Guidance Control and Dynamics},
  year = {1997},
  doi = {10.2514/2.4072},
}
Figure