RHONN Modelling-Enabled Nonlinear Predictive Control for Lateral Dynamics Stabilization of an In-Wheel Motor Driven Vehicle

Hao Chen, Junzhi Zhang, Chen Lv

TL;DR

This paper introduces a data-driven control strategy for stabilizing the lateral dynamics of in-wheel motor-driven vehicles (IMDVs). By replacing traditional physics-based models with a Recurrent High-Order Neural Network (RHONN), the researchers developed a Nonlinear Model Predictive Control (NMPC) system that adapts to complex, nonlinear driving conditions in real-time, significantly expanding the vehicle's stable handling limits.

What Problem Does This Paper Address?

Controlling the lateral stability (preventing skidding or spinning) of a vehicle is a major challenge, especially for in-wheel motor vehicles where each wheel can be controlled independently. Traditional controllers rely on physics-based models, like the "bicycle model."

While these models work well for normal driving, they fall apart during "limit handling"—situations with high lateral acceleration or slippery roads where tire behavior becomes highly nonlinear. To fix this, engineers often use complex empirical models like the "Magic Formula," but these require massive amounts of pre-calculated data and struggle to adapt to changing road conditions (like a sudden patch of ice). The authors aim to bridge this gap by using a neural network that "learns" the vehicle's physics on the fly.

Methods

The authors propose a hierarchical control architecture centered around two novel components:

  1. RHONN Modeling: Unlike standard Recurrent Neural Networks (RNNs) that use linear combinations of states, a Recurrent High-Order Neural Network (RHONN) uses high-order polynomials. This allows it to capture the complex, "curvy" relationships of tire dynamics more effectively. The model is updated in real-time using an Extended Kalman Filter (EKF), allowing it to adapt to different road friction levels without manual tuning.
  2. RHONN-based NMPC: The system uses the RHONN in two ways:
  3. Target Setting: A "neighbors-searching" algorithm uses the RHONN to find the optimal steady-state yaw rate and sideslip angle for the current conditions.
  4. Predictive Control: The NMPC uses the RHONN as its internal "crystal ball" to predict how the vehicle will react to different torque commands over a short future horizon, ultimately selecting the torque distribution that best maintains stability.

Datasets

The study relies on high-fidelity simulations rather than physical road tests.

  • CarSim/Simulink Co-simulation: The researchers used CarSim, an industry-standard vehicle dynamics simulator, to act as the "real" vehicle.
  • Scenarios: They evaluated the system using a Double-Lane Change (DLC) maneuver and Path Tracking on Slippery Curvatures.
  • Road Conditions: Tests were conducted on high-friction ($\mu=0.7$) and low-friction ($\mu=0.35$) surfaces to test the model's ability to handle nonlinear transitions.

Key Insights and Ideas

  • Data-Driven Physics: The RHONN acts as a "black box" that mimics physics. By using high-order interactions between neurons, it can approximate any dynamical system to a high degree of accuracy without needing the explicit equations for tire friction.
  • Avoiding "Aggressive" Targets: The "neighbors-searching" algorithm is a clever way to ensure the controller doesn't try to force the vehicle into an impossible state. It looks for the nearest stable equilibrium point, which prevents the controller from overreacting and causing further instability.
  • Torque Vectoring Advantage: Because IMDVs can apply different torques to each wheel (torque vectoring), they can generate a "yaw moment" (turning force) much faster and more precisely than traditional stability control systems that rely on hydraulic brakes.

Key Results

  • Model Accuracy: In low-friction tests, the RHONN model significantly outperformed traditional 7-Degree-of-Freedom (7-DoF) physics models. For longitudinal velocity estimation, the RHONN reduced the Root Mean Square Error (RMSE) by over 90% compared to linear models.
  • Higher Safe Speeds: In the double-lane change test on a slippery road, the RHONN-based NMPC remained stable at 74.4 km/h, whereas a standard Linear MPC (LMPC) failed at speeds above 62.8 km/h—an 18.4% improvement in the vehicle's stable operating envelope.
  • Computational Efficiency: Despite the complexity of the neural network, the average computation time was roughly 0.017 ms, well within the limits required for real-time automotive control (typically 20Hz to 100Hz).

Limitations and Future Work

  • Simulation vs. Reality: While CarSim is highly accurate, real-world sensors have noise and latency that were not fully explored here. Future work would need to validate this on a physical prototype.
  • Sensor Dependency: The system assumes it has accurate estimates of longitudinal and lateral velocity. In real vehicles, these are often estimated rather than measured directly, and errors in these estimates could degrade the neural network's performance.
  • Optimization Tuning: The performance of NMPC is highly dependent on weight coefficients ($q_i$ and $r_i$). The paper uses fixed weights, but an adaptive weighting strategy could further improve performance across a wider range of maneuvers.

Summary updated

Abstract

Featuring the fast response and flexibility in control allocation, an electric vehicle with in-wheel motors is a good platform for implementing advanced vehicle dynamics control. Among many active safety functions of an in-wheel motor driven vehicle (IMDV), lateral stability control is a key technology, which can be realized through torque vectoring. To further advance the lateral stabilization performance of the IMDV, in this article a novel data-driven nonlinear model predictive control (NMPC) is proposed based the recurrent high-order neural network (RHONN) modelling method. First, the new RHONN model is developed to represent vehicle’s nonlinear dynamic behaviors. Different from the conventional physics-based modelling method, the RHONN model forms high-order polynomials by neuron states to feature nonlinear dynamics. Based on the RHONN model, the steady-state responses of vehicle’s yaw rate and sideslip angle are iteratively optimized and set as the control objectives for low-level controller, aiming to improve the system robustness. Besides, a nonlinear model predictive controller is designed based on the RHONN, which is expected to improve the prediction accuracy during the receding horizon control. Further, a constrained optimization problem is formulated to derive the required yaw moment for vehicle lateral dynamics stabilization. Finally, the performance of the developed RHONN-based nonlinear MPC is validated on an IMDV in the CarSim/Simulink simulation environment. The validation results show that the developed approach outperforms the conventional method, and further improves the stable margin of the system. It is able to enhance the lateral stabilization performance of the IMDV under various driving scenarios, demonstrating the feasibility and effectiveness of the proposed approach.
BibTeX
@article{chen2022rhonn,
  author = {Hao Chen and Junzhi Zhang and Chen Lv},
  title = {{RHONN Modelling-Enabled Nonlinear Predictive Control for Lateral Dynamics Stabilization of an In-Wheel Motor Driven Vehicle}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2022},
  url = {https://openreview.net/forum?id=8TTge3XaLiv},
}

Loading related content...

Figure