Reduction of computational cost in driving simulation subsystems using approximation techniques

Abstract

Driving simulators are practical simulation tools in studying vehicle behavior and driver reaction in a safe and controllable condition. The development of a real time driving simulator evolves into complex highly integrated and interdependent systems that require vast amount of computer memory and computational time. This paper provides a study of employing approximation techniques in optimizing the computationally expensive simulation systems. Using the approximation techniques, a surrogate model can be constructed and used in the lieu of original codes. It can obviate the computational cost of highly integrated systems. A variety of approximation techniques can be used to simplify multidisciplinary simulations. In this paper, some well-known approximation techniques were reviewed including design of experiments, polynomial response surfaces, Kriging models and neural networks. A thorough review and study of various types of approximation techniques were made to construct efficient surrogate models for simulation subsystems. A surrogate assisted driving simulator (SADS) framework is then proposed that can significantly reduce the computational burden and achieve reasonable accuracy.