Validation of Tractor-semitrailer Vehicle Model based on Bayesian Hypothesis Testing

Abstract

The development of automated vehicles is the future of the automotive industry, and it has seen a rapid advance over the past decade. The design of automated vehicles relies on vehicle models, which predict their responses and control their behavior. This thesis presents a vehicle model validation framework to evaluate the validity of a simple (abstract) vehicle model against a complex (implementation) model. The aim is to quantify the validity of the abstract model and determine its suitability for an application. The validation procedure uses Bayesian hypothesis testing. First, two hypotheses are stated; the null hypothesis, which supports the abstract model’s validity, and the alternative, which rejects the model. Two approaches to Bayesian hypothesis testing are then studied. The first approach estimates the posterior distributions of the parameters of interest and evaluates whether the credible interval includes the null hypothesis. The second approach relies on Bayes factors, which compare the two hypotheses and indicate the most probable one. The methods presented are applied on a tractor-semitrailer combination under the driving context of city driving. The validation framework is evaluated for three cases of the abstract model. The first case refers to the calibrated model. The second and third cases concern an incorrectly tuned parameter, with double and half the calibrated value, respectively. The outcome is that the abstract model is deemed valid when calibrated, whereas it is invalid when a parameter is incorrectly tuned. The thesis also discusses the comparison between the two proposed validation approaches.