Computational optimization of a vehicle dynamic simulation model

The importance of simulation in motor-sport racing has steadily increased. Fine-tuning these simulators is complex and traditionally manual, relying on engineers’ expertise.This study develops a virtual driver and automates a parameter tuning pipeline for a race car simulator. The virtual driver based on the Data-enabled Predictive Control (DeePC) algorithm outperforms traditional Proportional Integral Derivative (PID) and Model Predictive Control (MPC) controllers and can potentially be extended to compute an optimal trajectory.A YAML-configured optimization pipeline with a custom Domain-Specific Language (DSL) was developed to accelerate parameter tuning. However, parameter interdependency is problematic for traditional optimization algorithms. Bayesian optimization can potentially find a global minimum despite interdependencies and converge to the optimum faster than traditional optimization methods.