⚡ electrical engineering

Path-Tracking Control for a Rear-Wheel-Drive Vehicle Governed by an Implicit ODE via State-Dependent Model Predictive Control

This paper proposes a State-Dependent Model Predictive Control (SDMPC) framework that exactly incorporates implicit vehicle dynamics and state-dependent drift equilibria into a single convex formulation, enabling robust path-tracking control for rear-wheel-drive vehicles both on friction-limited drifting manifolds and through complex maneuvers lacking equilibrium points.

Original authors: Javad Zolfaghari, A.S.M. Harithuddin, A.A. Nuraini, Raja Kamil, Reza Jazar

Published 2026-08-31

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Javad Zolfaghari, A.S.M. Harithuddin, A.A. Nuraini, Raja Kamil, Reza Jazar

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a car driving on a road that curves sharply. In normal driving, the tires grip the pavement firmly, and the car responds predictably to the steering wheel. But when a driver pushes a rear-wheel-drive car to its absolute limit, sliding sideways in a controlled drift, the physics change completely. The tires are no longer just gripping; they are sliding, and the forces they generate depend on how much weight is pressing down on them. That weight shifts constantly as the car accelerates or brakes. This creates a complex loop: the car's movement changes the weight on the tires, which changes the grip, which changes the movement again. For a computer trying to steer such a car, this loop is a mathematical nightmare. The equations that describe the car's motion do not simply tell the computer what will happen next; they hide the answer inside the question itself, requiring the computer to solve a puzzle where the solution is part of the problem.

This is the challenge researchers at Universiti Putra Malaysia and Tarleton State University set out to solve. They wanted to build a computer controller that could guide a rear-wheel-drive vehicle through difficult, high-speed drifts with the same precision a human expert might achieve, but without the risk of losing control. The difficulty lies in the fact that standard computer models used for self-driving cars often simplify the physics to make calculations faster. They assume the car behaves in a straight line or that the grip is constant. These simplifications work fine for gentle turns, but they fail spectacularly when a car is sliding sideways at the edge of traction. Other advanced methods exist that handle the full complexity, but they are so computationally heavy that they would require a supercomputer to run in real-time, making them useless for a car moving at speed. The researchers needed a middle ground: a system smart enough to understand the complex, shifting physics of a drifting car, but fast enough to run on the modest computer hardware found in a modern vehicle.

To achieve this, the team developed a new control strategy called State-Dependent Model Predictive Control. The core idea is to treat the car's complex, shifting behavior not as a mystery to be approximated, but as a set of rules that can be updated instantly. Instead of guessing how the car will behave based on a simple, fixed model, the computer looks at the car's current state—its speed, its angle, and how fast it is sliding—and recalculates the rules of physics for that exact moment. It does this by embedding the hidden, circular relationship between weight and grip directly into its calculations. By doing so, the computer avoids the errors that come from using simplified models. It creates a prediction of the car's future path that is mathematically exact for the current situation, without needing to solve a massive, time-consuming puzzle every time the steering wheel turns.

The researchers tested this system in two distinct scenarios to see how it held up. First, they asked the computer to guide the car along a perfect circle. In this test, the car had to maintain a steady drift at a specific angle and speed. The system successfully regulated the vehicle, keeping it on the circular path even when the car started from a position that was far off the mark. It corrected the errors smoothly, bringing the car back to the desired drift condition without jerking or losing stability. This proved that the new method could handle the complex physics of a steady drift, keeping the car balanced even when the initial conditions were quite severe.

The second test was far more demanding. The researchers programmed the car to drive a figure-eight pattern, a path that crosses over itself. This maneuver is particularly tricky because it requires the car to switch from drifting to the left to drifting to the right. As the car crosses the center of the figure-eight, the road curves straight for a split second before curving the other way. At that exact moment, there is no steady drifting state to hold onto; the car must transition through a point where the usual rules of a stable drift do not apply. Most control systems would struggle here, potentially losing the path or the drift entirely. However, the new controller managed the transition flawlessly. It guided the car through the zero-curvature point, smoothly reversing the direction of the slide, and kept the vehicle on the intended path with only tiny deviations. The car maintained its speed and grip, executing a full reversal of its sliding direction without ever losing control.

The results of these simulations show that this new approach offers a powerful solution for controlling vehicles at the very edge of their capabilities. It bridges the gap between simple, fast controllers that are too inaccurate for extreme driving, and complex, accurate controllers that are too slow to be practical. By solving the hidden mathematical loop of the car's physics directly, the system achieves high-precision control with a computational cost that is manageable for real-world use. While these results were achieved through computer simulations using a specific set of vehicle parameters, they demonstrate a clear path forward. The work suggests that it is possible to build autonomous systems capable of handling the most aggressive driving maneuvers, turning the chaotic physics of a high-speed drift into a predictable, controllable motion. The researchers note that future work will need to test these ideas on actual hardware and with different car models, but the foundation laid here shows that the mathematics of the drift can be tamed.

Technical Summary: Path-Tracking Control for a Rear-Wheel-Drive Vehicle via State-Dependent MPC

Problem Statement
Controlling rear-wheel-drive (RWD) vehicles operating near or beyond the tire friction limit presents two distinct structural challenges. First, the equations of motion are implicit: longitudinal load transfer causes the normal forces (which determine tire forces) to depend on the longitudinal acceleration, which is itself the state derivative. Consequently, the state derivative appears on the right-hand side of its own governing equation, preventing standard explicit integration without iterative solving. Second, the reference trajectory for drifting is not a fixed set-point but a state-dependent equilibrium manifold indexed by path curvature. Existing approaches often struggle with this: linear time-varying (LTV) MPC relies on first-order Taylor approximations that accumulate error over the prediction horizon, while full nonlinear MPC (NMPC) offers high accuracy but requires solving nonconvex optimization problems online, which is computationally prohibitive for embedded automotive hardware.

Methodology
The paper proposes a State-Dependent Model Predictive Control (SDMPC) framework that bridges the gap between the computational efficiency of linear MPC and the prediction accuracy of NMPC.

Exact SDC Formulation: The authors reformulate the implicit, nonlinear single-track vehicle model (incorporating Modified Elliptical Method (MEM) combined-slip tire forces, actuator dynamics, and Frenet-frame path kinematics) into an exact State-Dependent Coefficient (SDC) representation: $\dot{z} = A(z)z + B(z)U$ .
- To resolve the implicit ODE structure, the state vector is augmented to include the physical state derivatives ( $\dot{x}_1, \dot{x}_2, \dot{x}_3$ ) as explicit states.
- The algebraic coupling caused by load transfer is absorbed directly into the state-dependent matrices $A(z)$ and $B(z)$ via an exact factorization involving the term $1/(1 - f_{11})$ , where $f_{11}$ is the partial derivative of the longitudinal acceleration with respect to itself. This eliminates the need for Taylor-series residuals or fixed-point iterations.
- The resulting 10-state model is exact at every point in the augmented state space, not just a linearization about a specific operating point.
Receding-Horizon Control: The controller employs a successive-linearization approach where the SDC matrices are evaluated exactly at each predicted node along the horizon. This allows the construction of batch prediction matrices ( $\Phi$ and $\Gamma$ ) that encode the true nonlinear dynamics. The optimization problem remains a convex Quadratic Program (QP), ensuring computational tractability.
Two Control Formulations:
- Equilibrium Tracking: For constant-curvature paths, the controller regulates the augmented state to a fixed drift equilibrium where state derivatives are zero.
- Direct Output Tracking: For complex maneuvers where no single equilibrium exists (e.g., zero-curvature transitions), the controller is reformulated to drive a specific output vector (including sideslip angle, accelerations, and path errors) to zero, with the desired sideslip scheduled explicitly as a function of arc length.

Key Contributions

Implicit ODE Resolution: The paper presents the first known application of SDMPC to RWD drifting control that resolves the implicit nature of the equations of motion exactly within the SDC framework, avoiding linearization residuals.
Augmented State Representation: It introduces a 10-state SDC model that embeds the steady-state drifting condition ( $\dot{x}_i = 0$ ) directly into the dynamics, removing the need for separate offline equilibrium calculations.
Accuracy-Cost Trade-off: The study demonstrates that SDMPC achieves the prediction accuracy of full NMPC (by using exact SDC factorization) while retaining the convex QP structure of linear MPC, offering a favorable balance between accuracy and computational cost.
Generalized Tracking: The framework is extended from simple equilibrium stabilization to direct output tracking, enabling control of maneuvers involving sideslip sign reversals and zero-curvature transitions where no steady drift equilibrium exists.

Results
The controller was validated through two simulation studies:

Circular Path (Fixed Equilibrium): The controller successfully regulated the vehicle to a post-cusp drift equilibrium on a circular path. It demonstrated robustness across a "severity ladder" of initial conditions, including large lateral and course errors. The system recovered gracefully, with lateral errors converging to zero even when the initial deviation was significant or of opposite polarity.
Figure-Eight Path (Dynamic Output Tracking): The controller was tested on a self-intersecting figure-eight trajectory requiring a full sideslip-sign reversal through clothoid connectors (zero curvature). In this scenario, no single steady-state equilibrium exists. The output-tracking formulation successfully tracked the scheduled sideslip profile, managing the transition through zero curvature and reversing the sideslip sign twice per lap. Tracking errors remained small (lateral error < 2 cm on arcs, peaking at ~1.8 m during transitions) and recovered quickly.

Significance and Claims
The paper claims that the SDMPC architecture provides a unified solution for implicit, friction-limited vehicle dynamics. It asserts that a single exact prediction model can handle both equilibrium tracking (on constant curvature) and direct output tracking (on complex geometries with no equilibrium) within one architecture.

The authors explicitly qualify their claims:

The computational cost figures (SDMPC ~3.8 ms vs. NMPC ~50–200 ms) are indicative single-thread measurements on a desktop CPU, not hardware-in-the-loop benchmarks.
The stability results are empirical validations over tested scenarios; the paper does not provide a formal Lyapunov-based stability proof or recursive feasibility guarantee for the successive-linearization scheme on moving, curvature-indexed references.
The results are based on a specific vehicle parameter set and the MEM tire model; robustness to plant-model mismatch and experimental validation are identified as necessary future steps.

In summary, the work demonstrates that exact SDC factorization allows for high-fidelity prediction of implicit vehicle dynamics within a convex optimization framework, enabling robust path tracking in extreme drifting conditions without the computational burden of full nonlinear MPC.

Technical Summary: Path-Tracking Control for a Rear-Wheel-Drive Vehicle via State-Dependent MPC

More like this