Diffusion-Residual Model Predictive Steering Control for Vehicle Stabilization at the Limit of Handling

Jul 16, 2026

Upload by Ezra

A stabilizing model predictive controller (MPC) near the limit of handling must track a yaw-rate reference while enforcing a safe stable-handling envelope margin, both of which depend on the operating point and are unknown in advance. The objective is directional stabilization by bounding peak side-slip. A conditional diffusion residual model learns this uncertainty and conditions the controller’s reference and constraints on steering commands, producing a mean residual for reference resizing and a predictive spread for one-sided chance back-off. The resulting D-res MPC anticipates risk ahead of tracking error and improves behavior under low-friction divergence, meeting real-time constraints via offline tabulation and no in-loop diffusion.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

Abstract

Introduction

Limit of handling and stabilization objective

Model predictive control and calibration dependence

Model uncertainty and learning-based correction

Diffusion-Residual Model Predictive Steering Control for Vehicle Stabilization at the Limit of Handling under Model Uncertainty Bongsob Song arXiv :2607 . 10243v1 [ cs .RO] 11 Jul 2026 Abstract—At the limit of handling, a stabilizing model predictive controller (MPC) depends on the yaw-rate reference it tracks and the stable-handling envelope it enforces. Both the achievable reference and the safe envelope margin are operating-pointdependent and unknown a priori. Fixed or worst-case settings are therefore either too conservative or unsafe. The control objective is directional stabilization—bounding the side-slip angle to avoid loss of control—measured by peak side-slip. We learn this uncertainty with a conditional diffusion residual model and apply it to the controller’s reference and constraints rather than its control law. Conditioned on the steering command, the model returns the mean and a predictive spread of the nominalmodel residual. The mean re-sizes the tracked yaw reference; the spread, propagated over the prediction horizon, tightens the stable-handling envelope through a one-sided chance back-off. Together these form the proposed diffusion-residual MPC (D-res), so that caution is anticipated ahead of the tracking error rather than corrected after it by a high-gain loop. In our evaluation the mean re-size carries the bulk of the side-slip reduction andrecovers the low-friction divergence, while the second-moment back-off adds an anticipatory, self-gating envelope margin anda closed-loop-audited (aggregate) risk calibration that a meanonly reference leaves open. Because only two moments of the residual are required per command, the generator is tabulated offline and the online controller adds a single table lookup to the baseline MPC, running within the 100Hz budget on an embedded automotive processor (a measured worst-case 4.08ms per step, 41% of the 10ms budget, on an NVIDIA Jetson AGX Xavier), with no in-loop diffusion. Across a sevendegree-of-freedom (7-DOF) model and high-fidelity CarMaker co-simulation—an in-domain, simulation-scoped evaluation with the generator retrained per tier—spanning vehicle, tire, road, and maneuver diversity, D-res reduces peak side-slip where the fixed bicycle model is least accurate. It also restores directional stability on low-friction maneuvers, where the fixed reference over-commands the available grip. Index Terms—Model predictive control, diffusion models, vehicle stability control, limits of handling, chance-constrained control, real-time control, uncertainty quantification, active front steering. I. INTRODUCTION This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. The author is with the Department of Mobility Engineering, Ajou University, Suwon, South Korea (e-mail: [bsong@ajou.ac.kr](bsong@ajou.ac.kr)). Manuscript submitted July 8, 2026 . The companion code reproducing the paper’s main quantitative results will be made publicly available at [https://github.com/](https://github.com/)[ ](https://github.com/)[Vehicle-Intelligence-and-Control-Lab/DresMPC-matlab upon acceptance.](Vehicle-Intelligence-and-Control-Lab/DresMPC-matlab upon acceptance.) Fig. 1: Overview of the diffusion-residual MPC and its simulation data pipeline. T HE limit of handling is the regime vehicle operates close to the friction in which a road limit of its tires: the lateral-force characteristic saturates, the yaw dynamics turn nonlinear and become prone to divergence, and the margin for recovery collapses. It is the regime that decides the outcome of an emergency maneuver—an evasive lane change, a corner entered at excessive speed on a slippery road—and decades of active-safety engineering have been aimed at keeping the vehicle controllable there. Two actuator subsystems dominate. Active front steering (AFS) reshapes the yaw response through automatic corrections at

FAQs

What problem does the diffusion-residual MPC address at the limit of handling?

It addresses how MPC performance depends on an unknown yaw-rate reference and an uncertain safe-handling envelope margin, which vary with the operating point near tire friction limits.

How does the conditional diffusion residual model influence the controller?

Conditioned on steering command, it outputs a mean and a predictive spread of the nominal-model residual; the mean resizes the tracked yaw reference, while the spread enables one-sided chance back-off to tighten the envelope.

Why is the approach designed to avoid in-loop diffusion during real-time control?

Only two residual moments are required per command, so the generator is tabulated offline and the online controller performs a single table lookup, meeting embedded real-time budgets without running diffusion in the loop.

Jul 16, 2026

Upload by Ezra