Multi-Objective Optimization of a Path-following MPC for Vehicle Guidance: A Bayesian Optimization Approach

TL;DR

This paper presents a Bayesian optimization method to tune the cost functional parameters of a path-following model predictive control system for vehicles, enabling multi-objective optimization and Pareto-front computation.

Contribution

It introduces a Bayesian optimization approach for multi-objective tuning of MPC parameters in vehicle guidance, addressing the complexity of cost functional design.

Findings

Effective parameter tuning for vehicle MPC achieved

Pareto-fronts enable multi-objective trade-off analysis

Enhanced control performance through optimized cost functionals

Abstract

This paper tackles the multi-objective optimization of the cost functional of a path-following model predictive control for vehicle longitudinal and lateral control. While the inherent optimal character of the model predictive control and the direct consideration of constraints gives a very powerful tool for many applications, is the determination of an appropriate cost functional a non-trivial task. This results on the one hand from the number of degrees of freedom or the multitude of adjustable parameters and on the other hand from the coupling of these. To overcome this situation a Bayesian optimization procedure is present, which gives the possibility to determine optimal cost functional parameters for a given desire. Moreover, a Pareto-front for a whole set of possible configurations can be computed.