Research on the Update Method of Attitude Quaternion for Strapdown Inertial Navigation

Debiao Zhang, Jie Li, Yang Liu, Xiaokai Wei, Kaiqiang FengIEEE Access 2021 · DOI

Abstract

In strapdown inertial navigation systems (SINS), the accuracy of attitude and position determination is fundamentally linked to the quaternion update algorithm. This paper proposes a novel quaternion update method that directly solves the quaternion differential equation using third-order and fifth-order Taylor series approximations. For single and two-sample cases, the derived formulas align with traditional equivalent rotation vector (ERV) methods. For the three-sample case under coning motion, a fifth-order formula is developed and optimized using error analysis to minimize drift. Results indicate that this method achieves drift rate errors equivalent to optimized three-sample ERV algorithms but offers a more direct mathematical framework. To enhance computational efficiency, the fifth-order formula is simplified to a fourth-order version without compromising accuracy. Simulations and experimental car tests validate the theoretical robustness and practical applicability of the proposed approach.

Introduction

Accurate orientation, or attitude, is critical for the performance of Strapdown Inertial Navigation Systems (SINS) used in aerospace, autonomous vehicles, and robotics. These systems rely on high-frequency gyroscope data to track changes in orientation. Traditionally, this involves updating quaternions—four-dimensional mathematical entities that represent rotation without the risk of gimbal lock.

A common challenge in these calculations is the non-commutativity error, which arises during complex, high-frequency angular motions like "coning." While the equivalent rotation vector (ERV) method is the current industry standard for mitigating these errors, researchers at the North University of China have proposed a more direct approach using high-order Taylor series expansions to solve the quaternion differential equation.

The Challenge of Non-Commutativity

In 3D space, the sequence of rotations matters. When a vehicle undergoes simultaneous rotations about multiple axes, simple integration of angular velocity leads to a "drift" where the estimated orientation diverges from the true state. Most contemporary systems address this by calculating an ERV as an intermediate step. However, this paper argues that directly approximating the solution to the quaternion differential equation using Taylor series can capture complex motion dynamics more effectively, particularly in multi-sample scenarios.

Methodology: Taylor Series Expansion

The proposed method shifts the focus from the rotation vector back to the fundamental quaternion differential equation. The researchers derived update equations based on:

  1. High-Order Approximations: By utilizing 3rd-order and 5th-order Taylor series, the algorithm can better account for the curvature of motion between discrete data samples.
  2. Multi-Sample Optimization: Specific formulas were developed for single-sample, two-sample, and three-sample inputs. While the 1- and 2-sample versions converge with existing ERV methods, the 3-sample derivation allows for a more nuanced optimization of coefficients ($k_1$ and $k_2$) to minimize drift during coning motion.
  3. Computational Efficiency: Recognizing the constraints of real-time hardware, the authors simplified the 5th-order formula into a 4th-order implementation. This simplification maintains the optimized drift characteristics while reducing the processing overhead.

Experimental Validation

The efficacy of the Taylor series approach was tested through both synthetic simulations and real-world experiments:

  • Coning Motion Simulation: A standard benchmark involving rotation around an axis with a 1.5708° half-apex angle at 2 Hz. This represents a "worst-case" scenario for inertial drift.
  • Field Testing: A vehicle-mounted platform equipped with a custom miniature Inertial Measurement Unit (IMU). Ground truth was established using a high-precision LCI-1 strategic-level IMU and a Propak satellite receiver.

Key Findings

The research demonstrates several critical advantages:

  • Theoretical Consistency: The convergence of the Taylor series formulas with ERV methods for 1 and 2 samples validates the new framework as a robust generalization of existing techniques.
  • Precision in Complex Motion: In 3-sample simulations, the proposed method significantly outperformed standard 4th-order Runge-Kutta (RK4) methods, which are often used to solve quaternion or Bortz equations.
  • Stability: The optimized formula achieved a drift rate error of approximately $a\omega^7h^6 \times 10^{-5}$. In car tests, the method maintained tightly bounded errors in pitch, roll, and yaw, whereas RK4 methods exhibited divergent oscillations and offsets.

Conclusion and Future Directions

The Taylor series-based quaternion update method provides a mathematically direct and highly accurate alternative to traditional ERV-based navigation algorithms. By optimizing for multi-sample inputs, it effectively mitigates drift in high-frequency motion environments. Future work will explore the integration of this attitude algorithm into full navigation suites, focusing on its impact on position and velocity accuracy in GPS-denied environments and its performance under extreme, non-periodic vibrations.

Summary updated

BibTeX
@article{zhang2021research,
  author = {Debiao Zhang and Jie Li and Yang Liu and Xiaokai Wei and Kaiqiang Feng},
  title = {{Research on the Update Method of Attitude Quaternion for Strapdown Inertial Navigation}},
  journal = {IEEE Access},
  year = {2021},
  doi = {10.1109/access.2021.3058284},
}

Loading related content...

Figure