Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NeuroDOB: A Deep Neural Observer-Based Controller for Vehicle Lateral Dynamics

Sangmin Kim, Taehun Kim, Guntae Kim, Chang Mook Kang

TL;DR

NeuroDOB addresses the brittleness of model-based lateral control by replacing the conventional disturbance observer with a deep neural network that learns driver-specific steering compensation. By integrating a DNN-based compensation $\delta_c$ with a baseline LQR controller, the framework forms a stable, dual-system control that adapts to unmodeled dynamics and personalized driving styles, while preserving the stability guarantees of the LQR via a practical Lyapunov argument. Validation spans CarSim simulations with three road geometries and real-vehicle experiments on highway driving, showing substantial reductions in lateral tracking error (e.g., $e_y$ RMSE reductions up to ~86%) and strong generalization to unseen roads, though curvature-distribution shifts highlight the need for diverse training data. The work demonstrates the feasibility and impact of a data-driven observer that learns from surrogate-driver data to achieve personalized, adaptive lateral control in autonomous driving contexts, with practical implications for driver-vehicle-controller cooperation and safety.

Abstract

This paper proposes NeuroDOB, a deep neural network based observer controller for vehicle lateral dynamics, which replaces the conventional disturbance observer (DOB) with a deep neural network (DNN) to enhance personalized lateral control. Unlike conventional DOBs that compensate for general disturbances such as road friction variation and crosswind, NeuroDOB explicitly addresses unmodeled vehicle dynamics and driver-specific behaviors by learning the steering compensation signal from driver-in-the-loop simulations using CarSim's embedded controller as a surrogate driver. The proposed architecture integrates NeuroDOB with a linear quadratic regulator (LQR), where the DNN outputs a delta error correction added to the baseline LQR steering input to produce the final control command. Input features to the DNN include lateral position and yaw angle errors, and the LQR control input. Experimental validation using a lateral dynamic bicycle model within CarSim demonstrates that NeuroDOB effectively adapts to individual driving habits, improving lateral control performance beyond what conventional LQR controllers achieve. The results indicate the potential of deep neural network based observer to enable personalized and adaptive autonomous vehicle control. In cognitive terms, the proposed architecture can be viewed as a dual-system control structure. The baseline LQR corresponds to System 1, a model-based, fast, and analytic reasoning layer ensuring stability. The NeuroDOB acts as System 2, a reflective, data-driven layer that learns compensation from experience and corrects the analytical bias of System 1. Together, they form an integrated decision process analogous to human intuition-reflection interaction, enabling both stability and adaptability in lateral control.

NeuroDOB: A Deep Neural Observer-Based Controller for Vehicle Lateral Dynamics

TL;DR

NeuroDOB addresses the brittleness of model-based lateral control by replacing the conventional disturbance observer with a deep neural network that learns driver-specific steering compensation. By integrating a DNN-based compensation with a baseline LQR controller, the framework forms a stable, dual-system control that adapts to unmodeled dynamics and personalized driving styles, while preserving the stability guarantees of the LQR via a practical Lyapunov argument. Validation spans CarSim simulations with three road geometries and real-vehicle experiments on highway driving, showing substantial reductions in lateral tracking error (e.g., RMSE reductions up to ~86%) and strong generalization to unseen roads, though curvature-distribution shifts highlight the need for diverse training data. The work demonstrates the feasibility and impact of a data-driven observer that learns from surrogate-driver data to achieve personalized, adaptive lateral control in autonomous driving contexts, with practical implications for driver-vehicle-controller cooperation and safety.

Abstract

This paper proposes NeuroDOB, a deep neural network based observer controller for vehicle lateral dynamics, which replaces the conventional disturbance observer (DOB) with a deep neural network (DNN) to enhance personalized lateral control. Unlike conventional DOBs that compensate for general disturbances such as road friction variation and crosswind, NeuroDOB explicitly addresses unmodeled vehicle dynamics and driver-specific behaviors by learning the steering compensation signal from driver-in-the-loop simulations using CarSim's embedded controller as a surrogate driver. The proposed architecture integrates NeuroDOB with a linear quadratic regulator (LQR), where the DNN outputs a delta error correction added to the baseline LQR steering input to produce the final control command. Input features to the DNN include lateral position and yaw angle errors, and the LQR control input. Experimental validation using a lateral dynamic bicycle model within CarSim demonstrates that NeuroDOB effectively adapts to individual driving habits, improving lateral control performance beyond what conventional LQR controllers achieve. The results indicate the potential of deep neural network based observer to enable personalized and adaptive autonomous vehicle control. In cognitive terms, the proposed architecture can be viewed as a dual-system control structure. The baseline LQR corresponds to System 1, a model-based, fast, and analytic reasoning layer ensuring stability. The NeuroDOB acts as System 2, a reflective, data-driven layer that learns compensation from experience and corrects the analytical bias of System 1. Together, they form an integrated decision process analogous to human intuition-reflection interaction, enabling both stability and adaptability in lateral control.
Paper Structure (19 sections, 2 theorems, 34 equations, 16 figures, 6 tables)

This paper contains 19 sections, 2 theorems, 34 equations, 16 figures, 6 tables.

Table of Contents

  1. Introduction
  2. System Architecture and Modeling
  3. Lateral Bicycle Model
  4. LQR Controller Design
  5. Problem Formulation
  6. Proposed NeuroDOB Design
  7. Design and Theoretical Framework
  8. Training Objective and Optimization Strategy
  9. Comparison between Conventional DOB and NeuroDOB
  10. Simulation setup and Result
  11. Simulation setting
  12. Simulation results
  13. Simulation Case 1
  14. Simulation Case 2
  15. Simulation Case 3
  16. ...and 4 more sections

Key Result

Theorem 1

Consider the discrete-time vehicle lateral error dynamics: where $\delta_{\mathrm{LQR}}[k] = -K x[k]$ is the baseline LQR control law and $\delta_{\mathrm{c}}[k] = f_{\theta}(s[k])$ is the NeuroDOB compensation. If the learned compensation term $f_{\theta}(s[k])$ is bounded such that for some $\epsilon_1 > 0$, then the closed-loop system remains practically stable, i.e., where $\eta(\epsilon_1)

Figures (16)

  • Figure 1: Basic LQR and DOB system architecture.
  • Figure 2: Lateral dynamic bicycle model.
  • Figure 3: NeuroDOB architecture.
  • Figure 4: Architecture of the proposed NeuroDOB showing the data-driven compensation pathway integrated with the baseline LQR controller.
  • Figure 5: Road map shapes used in this study. Road Map 1 was used for training data collection and controller evaluation, Road Map 2 for independent testing, and Road Map 3 with modified geometry for additional training.
  • ...and 11 more figures

Theorems & Definitions (3)

  • Theorem 1
  • Corollary 1
  • Remark 1