Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Model-free Vehicle Rollover Prevention: A Data-driven Predictive Control Approach

Mohammad R. Hajidavalloo, Kaixiang Zhang, Vaibhav Srivastava, Zhaojian Li

TL;DR

This work targets rollover prevention for high-CG vehicles by eschewing explicit dynamic models in favor of Data-Enabled Predictive Control (DeePC). It introduces a reduced-dimension version (RD-DeePC) that leverages Hankel data and singular value decomposition to enable real-time, data-driven prediction and control. CarSim simulations with a sedan and a utility truck show that RD-DeePC outperforms Linear MPC (LMPC) in preventing rollovers while preserving maneuverability, even under riverbed-like disturbances, with the LTR constraint $|\text{LTR}|\le 1$ effectively enforced. The results indicate substantial computational gains from dimension reduction without sacrificing safety or performance, highlighting the practical potential of data-driven rollover protection in diverse driving conditions.

Abstract

Vehicle rollovers pose a significant safety risk and account for a disproportionately high number of fatalities in road accidents. This paper addresses the challenge of rollover prevention using Data-EnablEd Predictive Control (DeePC), a data-driven control strategy that directly leverages raw input-output data to maintain vehicle stability without requiring explicit system modeling. To enhance computational efficiency, we employ a reduced-dimension DeePC that utilizes singular value decomposition-based dimension reduction to significantly lower computation complexity without compromising control performance. This optimization enables real-time application in scenarios with high-dimensional data, making the approach more practical for deployment in real-world vehicles. The proposed approach is validated through high-fidelity CarSim simulations in both sedan and utility truck scenarios, demonstrating its versatility and ability to maintain vehicle stability under challenging driving conditions. Comparative results with Linear Model Predictive Control (LMPC) highlight the superior performance of DeePC in preventing rollovers while preserving maneuverability. The findings suggest that DeePC offers a robust and adaptable solution for rollover prevention, capable of handling varying road and vehicle conditions.

Model-free Vehicle Rollover Prevention: A Data-driven Predictive Control Approach

TL;DR

This work targets rollover prevention for high-CG vehicles by eschewing explicit dynamic models in favor of Data-Enabled Predictive Control (DeePC). It introduces a reduced-dimension version (RD-DeePC) that leverages Hankel data and singular value decomposition to enable real-time, data-driven prediction and control. CarSim simulations with a sedan and a utility truck show that RD-DeePC outperforms Linear MPC (LMPC) in preventing rollovers while preserving maneuverability, even under riverbed-like disturbances, with the LTR constraint effectively enforced. The results indicate substantial computational gains from dimension reduction without sacrificing safety or performance, highlighting the practical potential of data-driven rollover protection in diverse driving conditions.

Abstract

Vehicle rollovers pose a significant safety risk and account for a disproportionately high number of fatalities in road accidents. This paper addresses the challenge of rollover prevention using Data-EnablEd Predictive Control (DeePC), a data-driven control strategy that directly leverages raw input-output data to maintain vehicle stability without requiring explicit system modeling. To enhance computational efficiency, we employ a reduced-dimension DeePC that utilizes singular value decomposition-based dimension reduction to significantly lower computation complexity without compromising control performance. This optimization enables real-time application in scenarios with high-dimensional data, making the approach more practical for deployment in real-world vehicles. The proposed approach is validated through high-fidelity CarSim simulations in both sedan and utility truck scenarios, demonstrating its versatility and ability to maintain vehicle stability under challenging driving conditions. Comparative results with Linear Model Predictive Control (LMPC) highlight the superior performance of DeePC in preventing rollovers while preserving maneuverability. The findings suggest that DeePC offers a robust and adaptable solution for rollover prevention, capable of handling varying road and vehicle conditions.

Paper Structure

This paper contains 19 sections, 20 equations, 11 figures, 1 table.

Table of Contents

  1. Introduction
  2. Problem Formulation and Preliminaries
  3. Vehicle rollover prevention
  4. System dynamics and nonlinear MPC- based approach
  5. Linear MPC-based rollover prevention control
  6. Data Enabled Predictive Control (DeePC)
  7. Non-Parametric Representation of Linear Systems
  8. Data-EnablEd Predictive Control
  9. DeePC for rollover prevention
  10. CarSim Simulations
  11. Effectiveness of dimension reduction in DeePC
  12. Sedan vehicle
  13. Simulation setting
  14. Results
  15. Utility truck
  16. ...and 4 more sections

Figures (11)

  • Figure 1: Schematics of vehicle force diagram during turning.
  • Figure 2: Overview of the proposed framework using DeePC.
  • Figure 3: Comparison between DeePC and reduced-dimension DeePC
  • Figure 4: Road geometry used in CarSim.
  • Figure 5: A snapshot of the CarSim environment illustrating the implementation of DeePC1 and LMPC1. The image captures the moment at $t=21.7s$, where the green vehicle (LMPC1) violates the LTR constraint and begins to lose stability, while the blue vehicle (DeePC1) successfully navigates the sharp turn.
  • ...and 6 more figures

Theorems & Definitions (3)

  • Definition 1
  • Remark 1: Dimension of $g$
  • Remark 2: Practical selection of $T$