Performance Analysis of Quaternion vs Euler-based Approaches in Reduced INS/GNSS Fusion Techniques

Unmanned vehicle navigation using various combination of sensors in different fields, civilian or military, has always known to be a challenging and nontrivial task. Bayesian filter algorithms are usually being used for combining and optimally processing measurements provided by such sensors, considering the available knowledge such as the uncertainty in measurements. Since the performance of any navigation system mainly depends on the efficiency of the chosen filter algorithm associated with the utilized sensors’ quality, the choice of suitable filtering approach (according to the demand application) can be considered as a critical and essential designing point. In this paper, the performance of two different attitudes representation filtering approaches (Quaternion and Euler angles-based) is investigated, when applied in Multiplicative Extended Kalman Filter (MEKF) design. The overall performance and accuracy of both approaches were analyzed and compared during various GNSS signal outages for a self-driving cars application in straight and curved trajectory paths. Multiple in-filed experiments were conducted using a special setup that includes a MEMS-based SPATIAL Advanced Navigation INS/GNSS system where the output of both approaches was compared to the ground truth of a utilized reference navigation system. The comparison includes 1) Position, Velocity and Attitude (PVA) estimation accuracy, 2) processing time and computational complexity. Results concluded that each approach has its own merits in demerits where the Quaternion-based approach outperforms the Euler one in terms of PVA navigation solution estimation accuracy. On the other hand, a faster calculation for the Euler-based approach associated with lower computational complexity was significantly noticeable.