What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance?
Model predictive control (MPC) is a promising approach to the lateral and longitudinal control of autonomous vehicles. However, the parameterization of the MPC with respect to high-level requirements such as passenger comfort, as well as lateral and longitudinal tracking, is challenging. Numerous tu...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-01-01
|
| Series: | Mathematics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2227-7390/11/2/465 |
| _version_ | 1797439034455228416 |
|---|---|
| author | David Stenger Robert Ritschel Felix Krabbes Rick Voßwinkel Hendrik Richter |
| author_facet | David Stenger Robert Ritschel Felix Krabbes Rick Voßwinkel Hendrik Richter |
| author_sort | David Stenger |
| collection | DOAJ |
| description | Model predictive control (MPC) is a promising approach to the lateral and longitudinal control of autonomous vehicles. However, the parameterization of the MPC with respect to high-level requirements such as passenger comfort, as well as lateral and longitudinal tracking, is challenging. Numerous tuning parameters and conflicting requirements need to be considered. In this paper, we formulate the MPC tuning task as a multi-objective optimization problem. Its solution is demanding for two reasons: First, MPC-parameterizations are evaluated in a computationally expensive simulation environment. As a result, the optimization algorithm needs to be as sample-efficient as possible. Second, for some poor parameterizations, the simulation cannot be completed; therefore, useful objective function values are not available (for instance, learning with crash constraints). In this work, we compare the sample efficiency of multi-objective particle swarm optimization (MOPSO), a genetic algorithm (NSGA-II), and multiple versions of Bayesian optimization (BO). We extend BO by introducing an adaptive batch size to limit the computational overhead. In addition, we devise a method to deal with crash constraints. The results show that BO works best for a small budget, NSGA-II is best for medium budgets, and none of the evaluated optimizers are superior to random search for large budgets. Both proposed BO extensions are, therefore, shown to be beneficial. |
| first_indexed | 2024-03-09T11:45:58Z |
| format | Article |
| id | doaj.art-ee5d391ca1d0410581ec1a7d55c7d7dc |
| institution | Directory Open Access Journal |
| issn | 2227-7390 |
| language | English |
| last_indexed | 2024-03-09T11:45:58Z |
| publishDate | 2023-01-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Mathematics |
| spelling | doaj.art-ee5d391ca1d0410581ec1a7d55c7d7dc2023-11-30T23:22:36ZengMDPI AGMathematics2227-73902023-01-0111246510.3390/math11020465What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance?David Stenger0Robert Ritschel1Felix Krabbes2Rick Voßwinkel3Hendrik Richter4Institute of Automatic Control (IRT), RWTH Aachen University, D-52074 Aachen, GermanyDepartment Automated Driving Functions, IAV GmbH, D-09120 Chemnitz, GermanyFaculty of Automotive Engineering, Zwickau University of Applied Sciences, D-08056 Zwickau, GermanyFaculty of Automotive Engineering, Zwickau University of Applied Sciences, D-08056 Zwickau, GermanyFaculty of Engineering, HTWK Leipzig University of Applied Sciences, D-04277 Leipzig, GermanyModel predictive control (MPC) is a promising approach to the lateral and longitudinal control of autonomous vehicles. However, the parameterization of the MPC with respect to high-level requirements such as passenger comfort, as well as lateral and longitudinal tracking, is challenging. Numerous tuning parameters and conflicting requirements need to be considered. In this paper, we formulate the MPC tuning task as a multi-objective optimization problem. Its solution is demanding for two reasons: First, MPC-parameterizations are evaluated in a computationally expensive simulation environment. As a result, the optimization algorithm needs to be as sample-efficient as possible. Second, for some poor parameterizations, the simulation cannot be completed; therefore, useful objective function values are not available (for instance, learning with crash constraints). In this work, we compare the sample efficiency of multi-objective particle swarm optimization (MOPSO), a genetic algorithm (NSGA-II), and multiple versions of Bayesian optimization (BO). We extend BO by introducing an adaptive batch size to limit the computational overhead. In addition, we devise a method to deal with crash constraints. The results show that BO works best for a small budget, NSGA-II is best for medium budgets, and none of the evaluated optimizers are superior to random search for large budgets. Both proposed BO extensions are, therefore, shown to be beneficial.https://www.mdpi.com/2227-7390/11/2/465Bayesian optimizationmetaheuristicsmodel predictive controlmulti-objective optimizationcontroller tuningvehicle guidance |
| spellingShingle | David Stenger Robert Ritschel Felix Krabbes Rick Voßwinkel Hendrik Richter What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance? Mathematics Bayesian optimization metaheuristics model predictive control multi-objective optimization controller tuning vehicle guidance |
| title | What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance? |
| title_full | What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance? |
| title_fullStr | What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance? |
| title_full_unstemmed | What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance? |
| title_short | What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance? |
| title_sort | what is the best way to optimally parameterize the mpc cost function for vehicle guidance |
| topic | Bayesian optimization metaheuristics model predictive control multi-objective optimization controller tuning vehicle guidance |
| url | https://www.mdpi.com/2227-7390/11/2/465 |
| work_keys_str_mv | AT davidstenger whatisthebestwaytooptimallyparameterizethempccostfunctionforvehicleguidance AT robertritschel whatisthebestwaytooptimallyparameterizethempccostfunctionforvehicleguidance AT felixkrabbes whatisthebestwaytooptimallyparameterizethempccostfunctionforvehicleguidance AT rickvoßwinkel whatisthebestwaytooptimallyparameterizethempccostfunctionforvehicleguidance AT hendrikrichter whatisthebestwaytooptimallyparameterizethempccostfunctionforvehicleguidance |