Clausen et al. Table 2. Parameter Values of the Models for Each Iteration Seen in Figure 6 WN QN GM #1 GM #2 GM #3 Navchip datasheet 6.85e-7 - 5.00e-1 - - 1st iteration Figure 6(a) 7.66e-7 - - 2nd iteration Figure 6(b) 7.52e-7 8.13e-4 1.49e-1 - 2.91e-8 1.10e-8 3rd iteration Figure 6(c) 7.36e-7 1.34e-2 3.34e-4 6.36e-1 9.93e-9 3.42e-8 7.77e-9 XSENS MTi-G datasheet 7.62e-5 2.50e-2 - - 1st iteration Model (a) 3.05e-6 - - 2nd iteration Model (b) 2.65e-6 1.52e-3 2.06e+2 - 2.29e-7 4.62e-5 3rd iteration Model (c) 2.03-06 5.94e-2 9.43e-4 2.08e+2 1.15e-7 2.05e-7 4.68e-5 2.35e-9 2.25e-6 5.05e-3 3.45e-8 2.26e-6 2.27e-6 - 9.40e-9 - 2.07e+2 a criterion to select the models that better predict the WV of future replicates of this error signal. Although the platform allows the computation of this criterion for each individual model, we need this to compare it to other models so as to understand which of these has a reasonably low WVIC value. To do so, the GUI provides the button Reduce Model Automatically in which it is possible to define an overall model that includes all the others. All possible combinations of the stochastic models are checked automatically; the results of the best model-combination in terms of the WVIC is plotted directly in the GUI with its parameters. Using the additional information coming from the model selection process, it is, therefore, possible to make a final decision on which model to use within the navigation filter. 4.58e-5 - - NOTE: The upper half describes the Navchip IMU, whereas the lower half depicts the values for the XSENS MTi-G IMU. IMPACT ON NAVIGATION A precise knowledge of the IMU sensor stochastic properties improves the navigation solution since the fusion filter has to be tuned with those values. The better the tuning, the better the navigation solution in different applications [13], [14]. Figure 8 shows an example of a successful sensor Figure 8. Example of correct modeling of an IMU used for position estimation of a car-trajectory [13]. (a) Horizontal trajectory. (b) Planimetric and altimetric errors. AUGUST 2018 IEEE A&E SYSTEMS MAGAZINE 35