Application of Data-Driven Model Predictive Control for Autonomous Vehicle Steering

Jiarui Zhang, Aijing Kong, Yu Tang, Zhichao Lv, Lulu Guo, Peng Hang

Abstract

With the development of autonomous driving technology, there are increasing demands for vehicle control, and MPC has become a widely researched topic in both industry and academia. Existing MPC control methods based on vehicle kinematics or dynamics have challenges such as difficult modeling, numerous parameters, strong nonlinearity, and high computational cost. To address these issues, this paper adapts an existing Data-driven MPC control method and applies it to autonomous vehicle steering control. This method avoids the need for complex vehicle system modeling and achieves trajectory tracking with relatively low computational time and small errors. We validate the control effectiveness of the algorithm in specific scenario through CarSim-Simulink simulation and perform comparative analysis with PID and vehicle kinematics MPC, confirming the feasibility and superiority of it for vehicle steering control.

Introduction

Model Predictive Control (MPC) is a cornerstone of autonomous vehicle research due to its ability to handle multi-variable constraints and optimize future behavior. However, traditional MPC implementations are often "model-heavy." They typically rely on either a kinematic model—which is computationally simple but loses accuracy at higher speeds—or a dynamic model, which is accurate but involves complex nonlinear equations and numerous parameters that are difficult to calibrate in real-time.

This creates a "complexity vs. performance" trade-off: simple models lead to poor control precision, while complex models demand high computational power, potentially causing latency in critical driving scenarios. This paper explores a Data-driven Model Predictive Control (DDMPC) approach that bypasses traditional mathematical modeling by learning system behavior directly from historical input-output data.

Methodology: The Data-Driven Framework

The core of this approach is the application of Willems' Lemma, a fundamental result in behavioral systems theory. Instead of using differential equations to describe vehicle physics, the system's dynamics are represented implicitly through a Hankel Matrix.

Workflow of the proposed Data-driven MPC approach.

The process involves three main stages:

  1. Data Acquisition and Preprocessing: Open-loop steering data (input) and vehicle coordinates/heading (output) are collected. This data must be "persistently exciting" to ensure the Hankel matrix captures the full range of vehicle dynamics.
  2. Hankel Matrix Construction: The preprocessed data is organized into matrices that represent the system's behavior. If the input sequence is sufficiently rich, any future trajectory can be expressed as a linear combination of the columns in these matrices.
  3. Optimization and Constraints: The DDMPC framework minimizes a cost function that balances trajectory tracking error against steering effort. Constraints are applied to ensure safety and to anchor the prediction to the vehicle's current state using recent history.

Experimental Setup

The algorithm was validated using a high-fidelity simulation environment:

  • Simulation Tools: CarSim was used to model a D-Class Sedan, integrated with MATLAB/Simulink for the control logic.
  • Data Collection: An initial dataset of $N=646$ points was generated through open-loop maneuvers.
  • Test Scenario: A dual-lane switching maneuver conducted at a constant speed of 36 km/h, designed to test the controller's ability to handle rapid lateral transitions.

Key Results and Discussion

The DDMPC method was compared against a standard PID controller and a traditional Kinematics-based MPC. The results highlighted several key advantages of the data-driven approach.

Trajectory Tracking Accuracy

As shown in the comparison below, the DDMPC maintained superior tracking performance, particularly during the transitions between curved and straight path segments.

Comparison of trajectory tracking performance between DDMPC, PID, and Kinematics MPC.

While PID and Kinematic MPC exhibited noticeable "overshoot" or lag during lane-change maneuvers, the DDMPC remained closely aligned with the reference line. Statistical analysis showed that DDMPC trajectory errors were generally confined between -0.1m and 0.2m, with significantly fewer outliers than the competing methods.

Computational Efficiency and Stability

One of the most significant findings was the reduction in computational overhead. Because the DDMPC avoids solving complex nonlinear physics equations, its computation time was nearly 50% lower than that of the Kinematics MPC. Furthermore, the steering output remained smooth and within a stable range, which is essential for passenger comfort and vehicle stability.

Conclusion and Future Work

This study demonstrates that a data-driven approach can effectively control autonomous vehicle steering without the need for explicit physical modeling. By leveraging historical data, DDMPC achieves higher accuracy and lower computational costs than traditional methods. Future research will focus on testing the algorithm under variable speeds and updating the Hankel matrix online to account for changing road conditions.

Summary updated

Abstract

With the development of autonomous driving technology, there are increasing demands for vehicle control, and MPC has become a widely researched topic in both industry and academia. Existing MPC control methods based on vehicle kinematics or dynamics have challenges such as difficult modeling, numerous parameters, strong nonlinearity, and high computational cost. To address these issues, this paper adapts an existing Data-driven MPC control method and applies it to autonomous vehicle steering control. This method avoids the need for complex vehicle system modeling and achieves trajectory tracking with relatively low computational time and small errors. We validate the control effectiveness of the algorithm in specific scenario through CarSim-Simulink simulation and perform comparative analysis with PID and vehicle kinematics MPC, confirming the feasibility and superiority of it for vehicle steering control.
BibTeX
@misc{zhang2024application,
  author = {Jiarui Zhang and Aijing Kong and Yu Tang and Zhichao Lv and Lulu Guo and Peng Hang},
  title = {{Application of Data-Driven Model Predictive Control for Autonomous Vehicle Steering}},
  year = {2024},
  eprint = {2407.08401},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2407.08401},
}
Figure