Computing uncertainty measures of location estimates for autonomous vehicles

Determining the location of an autonomous vehicle is an important problem in navigating the vehicle in an unstructured environment. The vehicle controller estimates the vehicle's location and calculates the covariance matrix as an uncertainty measure for the location estimate. The real-time implementation of the controller makes the calculation of the covariance matrix an important issue. There is no exact method for calculating the covariance matrix because of the nonlinear nature of the location estimator. Approximations are needed. In this article, several approximation methods are compared through simulation. The comparisons are focused on each incremental change and on the cummulative effects of the trajectory-following of a path. The robustness of approximations is also studied by investigating the behavior of approximations under different distributional assumptions for measurement models. The results are useful for finding the scopes and limits of applicability of the approximation