19.6 Performance Comparison Analysis

The performance analysis of all estimation filters presented in Section 19.4 was conducted using the synthetic test data described in Section 19.5. The performance of the UKF using zero-, first-, second-, and third-order dynamic models is compared against the NLLSQ and also against the UGPF using the second-order dynamic model. The performance metric used in all cases is the root mean squared (RMS) error. The performance results to be presented below indicate that the second-order UKF was the best filter to use.

Figures 19.4 through 19.6 show the results the estimated track results of a single run of the UKF filter using the second -order model compared against synthetic truth data. Observe that the rotational rates and accelerations shown in Figures 19.5 and 19.6 are in body-referenced coordinates. Figure 19.7 shows a close-up of a small section of the lateral position truth trajectory with the estimated trajectories from multiple solvers superimposed over the truth. Identical noisy observations and initialization were used as input to all estimation methods. Analysis of this data leads to two conclusions: first, we note that the NLLSQ and the UKF with a zeroth-order dynamic model generate nearly identical estimates. Since the zeroth-order model is a constant position model and the NLLSQ assumes a position-only state vector, the similarity in performance of the NLLSQ and UKF(zeroth order) is to be expected. The second conclusion from Figure ...

Get Bayesian Estimation and Tracking: A Practical Guide now with the O’Reilly learning platform.

O’Reilly members experience live online training, plus books, videos, and digital content from nearly 200 publishers.