TL;DR
This paper tackles the confusion caused by two different quaternion multiplication conventions (Hamilton's and a "flipped" version) prevalent in robotics and aerospace. The authors propose discontinuing the flipped version in favor of a more consistent approach using Hamilton's multiplication with a specific matrix-to-quaternion mapping, offering practical recipes for detection and migration.
What Problem Does This Paper Address?
The paper addresses a significant practical problem in fields like robotics and aerospace: the existence of two common quaternion multiplication conventions. While quaternions are powerful for representing rotations (e.g., spacecraft attitude), the use of both Hamilton's original multiplication and a "flipped" version (often associated with NASA's Jet Propulsion Laboratory) leads to confusion, errors, and difficulties in exchanging formulas and implementations. This ambiguity arises from how quaternions are mapped to rotation matrices, particularly in the "passive world-to-body" convention, where a mismatch can occur if not handled carefully. The authors argue this issue is avoidable and stems from a lack of understanding.
Methods
The authors propose an alternative solution that involves using a different correspondence rule between quaternions and rotation matrices. Instead of the "flipped" multiplication, they advocate for a mapping that maintains homomorphy with Hamilton's original quaternion multiplication. This approach, they argue, leads to greater formal similarity with matrix representations of rotations and avoids the need for the flipped multiplication. The paper also provides practical "recipes" for:
- Detecting which quaternion multiplication convention is being used in existing formulas or software.
- Migrating formulas and algorithms from one convention to another, offering both "translate" and "interface" procedures.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- The "flipped" quaternion multiplication is unnecessary: The authors argue that the flipped multiplication, introduced to resolve a perceived issue with passive world-to-body rotations, is not fundamentally required.
- A more consistent approach exists: By adopting a specific quaternion-to-matrix mapping (CH) that is homomorphic with Hamilton's original multiplication, the need for the flipped version is eliminated.
- Formal similarity matters: The proposed approach (Hamilton's multiplication with CH mapping) exhibits greater formal similarity to matrix representations of rotations, especially when dealing with concepts like angular velocity and rotation vectors. This similarity reduces the likelihood of errors.
- The problem is historical and convention-driven: The confusion stems from how quaternions are mapped to matrices and how this interacts with the multiplication operation, rather than an inherent limitation of quaternions themselves.
- Practical tools are provided: The paper offers concrete methods for identifying which convention is in use and for converting existing code and formulas, making the transition easier.
Key Results
The paper doesn't present quantitative results in terms of accuracy or performance benchmarks. Instead, its key results are:
- A clear argument for discontinuing the flipped quaternion multiplication.
- The proposal of an alternative, more consistent convention using Hamilton's multiplication and a specific matrix-to-quaternion mapping (CH).
- Demonstration of greater formal similarity between this proposed convention and matrix representations of rotations, particularly for concepts like angular velocity and rotation vectors.
- Development of practical recipes for detecting and migrating between the two quaternion multiplication conventions.
Limitations and Future Work
- Limited scope of "recipes": While the recipes are presented as efficient, their application might require careful decomposition of complex tools into smaller components.
- Focus on homomorphic conventions: The paper primarily focuses on comparing homomorphic quaternion-matrix conventions, acknowledging that antihomomorphic ones are less common and potentially problematic.
- Future work: The authors suggest that further investigation into the "rotation range intuition" (RRI) could be beneficial, although they propose using angle-axis parameters as a more intuitive alternative for this purpose. They also imply that the community should converge on a single, consistent convention to reduce ongoing costs.
Summary updated