Tongji University Researchers Unveil Advanced Control Strategy for Electric Vehicle Stability and Precision
In the rapidly evolving world of autonomous driving, ensuring both safety and precision during high-speed maneuvers remains one of the most pressing engineering challenges. A groundbreaking study from Tongji University has introduced a novel hierarchical control framework designed to significantly enhance the trajectory tracking accuracy and yaw stability of four-wheel-independent-drive electric vehicles (4WID-EVs), particularly under extreme driving conditions such as emergency obstacle avoidance.
The research, led by Dr. Shuping Chen, Professor Zhiguo Zhao, and Kun Zhao from the School of Automotive Studies at Tongji University, was recently published in the Journal of Tongji University (Natural Science). Their work presents a sophisticated, multi-layered control strategy that seamlessly integrates path following with vehicle dynamics stability, addressing a critical gap in current autonomous vehicle control systems.
As autonomous vehicles transition from controlled environments to complex real-world scenarios, their ability to handle sudden, high-risk situations—like swerving to avoid a collision—becomes paramount. Traditional control systems often prioritize path tracking accuracy, sometimes at the expense of vehicle stability. This trade-off can lead to dangerous outcomes, including loss of control, skidding, or even rollover, especially on low-friction surfaces or at high speeds. The Tongji team’s approach directly confronts this issue by unifying trajectory tracking and yaw stability within a single, cohesive control architecture.
The core of their innovation lies in a two-tiered control system. At the upper level, the researchers employ a Linear Time-Varying Model Predictive Control (LTV MPC) algorithm. This advanced control method is renowned for its ability to predict future vehicle behavior and optimize control inputs over a finite time horizon, taking into account various system constraints. Unlike many existing approaches that rely on simplified vehicle models, the Tongji team’s LTV MPC is built upon an 8-degree-of-freedom (8-DOF) vehicle dynamics model. This higher-fidelity model captures critical dynamics such as longitudinal, lateral, yaw, and roll motions, providing a more accurate representation of the vehicle’s behavior, especially during aggressive maneuvers.
A key differentiator of this upper-layer controller is its integration of a PID-based speed tracking mechanism directly into the MPC optimization loop. This integration allows the controller to dynamically account for changes in longitudinal velocity during the prediction phase. In real-world driving, vehicle speed is rarely constant, particularly during emergency maneuvers where rapid acceleration or braking may occur. By embedding the longitudinal speed control within the MPC framework, the system can generate a total required torque command that is inherently synchronized with the lateral and yaw stability objectives. This holistic approach ensures that the vehicle’s longitudinal and lateral dynamics are not treated as separate problems but are instead optimized in concert, leading to smoother and more stable overall performance.
The LTV MPC controller calculates two primary control outputs: the desired front wheel steering angle and an additional yaw moment. The steering angle is used to guide the vehicle along the intended path, while the additional yaw moment is a crucial element for maintaining stability. This moment is not generated by traditional mechanical means but is instead produced through the intelligent distribution of torque to the four independent wheels—a capability unique to 4WID-EVs.
This is where the lower layer of the control system comes into play. The second tier of the hierarchy is responsible for control allocation, a process that determines how the total driving or braking torque and the additional yaw moment are distributed among the four wheels. The Tongji researchers utilize a Quadratic Programming (QP) method to solve this allocation problem. QP is a powerful optimization technique that can efficiently find the best solution while respecting multiple physical and operational constraints.
The optimization objectives in the QP problem are multifaceted, reflecting the complex trade-offs involved in vehicle control. The first objective is to minimize the torque tracking error, ensuring that the actual torque delivered by each wheel closely matches the command from the upper layer. The second objective is to minimize tire utilization. Tire utilization is a measure of how close a tire is operating to its physical friction limit. By keeping this utilization low, the control system preserves a safety margin, allowing the tires to generate additional lateral or longitudinal forces if needed, which is essential for maintaining stability. The third objective is to minimize longitudinal slip energy loss, which helps to prevent excessive wheel slip and improves energy efficiency.
The QP solver operates under several critical constraints. These include the maximum and minimum torque output of each electric motor, which are dictated by the hardware limitations of the drivetrain. Equally important are the road friction constraints, which ensure that the combined longitudinal and lateral forces at each tire do not exceed the available friction ellipse, a fundamental limit imposed by the road surface and tire condition. The controller uses a combined-slip brush tire model to accurately represent the nonlinear coupling between longitudinal and lateral tire forces, a feature that is vital for realistic simulation and control in extreme conditions.
To validate the effectiveness of their proposed control strategy, the research team conducted extensive simulations using a high-fidelity 14-degree-of-freedom (14-DOF) vehicle model. This detailed model, which includes individual wheel dynamics and suspension travel, serves as the “plant” or virtual representation of the real vehicle, providing a more realistic testbed than the 8-DOF prediction model. The primary test scenario was a double-lane-change maneuver, a standard benchmark for evaluating a vehicle’s emergency handling and stability. This maneuver simulates the rapid swerving required to avoid an obstacle in the road, placing significant stress on the vehicle’s control systems.
The simulation results were compelling. The researchers tested the controller across a range of speeds, from 36 km/h to 90 km/h, on a high-friction surface (µ = 0.85). At all speeds, the vehicle demonstrated excellent trajectory tracking, with the maximum lateral deviation remaining under 0.28 meters. While, as expected, the tracking error increased slightly with speed due to the greater dynamic forces involved, the performance remained well within acceptable limits. More importantly, the vehicle’s yaw rate and sideslip angle—the key indicators of stability—remained within safe bounds throughout the maneuver. The controller effectively used the additional yaw moment, generated by creating a torque difference between the left and right wheels, to keep the vehicle stable and on course.
The robustness of the control strategy was further demonstrated in simulations on different road surfaces. The team tested the system at a constant speed of 50 km/h on both a high-friction surface (µ = 0.8) and a low-friction surface (µ = 0.3), which could represent wet or icy roads. On the high-friction surface, the vehicle tracked the path with exceptional precision. On the low-friction surface, as anticipated, the tracking performance was more challenging, with a slight increase in lateral error and a small delay in response. However, the maximum lateral deviation was still less than 0.12 meters, and the vehicle maintained stability without any signs of losing control. This result is particularly significant, as it shows the controller’s ability to adapt to changing road conditions, a critical capability for real-world deployment.
Perhaps the most telling validation came from a direct comparison between the proposed stability-coordinated controller and a simpler trajectory-tracking-only controller. In a simulation at 72 km/h on a low-friction surface (µ = 0.4), the difference was stark. The stability-coordinated controller achieved a smaller lateral tracking error and a faster, more accurate speed response. More importantly, the vehicle’s sideslip angle and yaw rate were significantly smaller and more stable when the coordinated control was active. This demonstrates that the additional yaw moment is not just a theoretical construct but a practical tool that actively improves both stability and tracking precision. The controller achieved a better path by making the vehicle more stable, effectively proving that stability and accuracy are not competing goals but can be mutually reinforcing.
The implications of this research extend beyond academic interest. For the automotive industry, this work provides a clear blueprint for developing next-generation vehicle stability control systems. As 4WID-EV platforms become more common, particularly in high-performance and autonomous vehicles, the ability to leverage individual wheel torque for stability will be a key differentiator. The Tongji team’s control strategy offers a practical and effective way to unlock this potential.
Furthermore, the integration of a high-fidelity prediction model and a multi-objective optimization framework sets a new standard for control system design. It moves away from the traditional, siloed approach to vehicle control—where steering, braking, and propulsion are managed by separate systems—and towards a holistic, integrated approach. This is essential for the future of autonomous driving, where the vehicle must act as a single, intelligent agent, capable of making complex, real-time decisions that balance multiple, often competing, objectives.
The research also highlights the importance of model fidelity. By using an 8-DOF model for prediction and a 14-DOF model for simulation, the team ensured that their control strategy was tested under realistic conditions. This attention to detail increases the confidence that the algorithm will perform well when implemented on a real vehicle, a critical step in the development process.
Looking ahead, the authors have identified several promising avenues for future research. One is the integration of roll stability into the control framework. While the current work focuses on yaw stability, a vehicle’s roll dynamics are equally important, especially during high-lateral-acceleration maneuvers. A control system that can simultaneously manage yaw and roll would provide an even higher level of safety. Another area is the development of adaptive methods for tuning the weighting factors in the QP optimization. These factors determine the relative importance of torque tracking, tire utilization, and energy loss. An adaptive system that can adjust these weights in real-time based on driving conditions could further optimize performance.
In conclusion, the work by Chen, Zhao, and Zhao represents a significant advancement in the field of vehicle dynamics and control. Their hierarchical, model-predictive control strategy for 4WID-EVs successfully bridges the gap between precise path tracking and robust vehicle stability. Through rigorous simulation and a well-designed control architecture, they have demonstrated a system that performs reliably across a wide range of speeds and road conditions. This research not only contributes to the academic body of knowledge but also provides a practical solution with the potential to make future autonomous vehicles safer and more capable.
Shuping Chen, Zhiguo Zhao, Kun Zhao, School of Automotive Studies, Tongji University, Journal of Tongji University (Natural Science), DOI:10.11908/j.issn.0253-374x.24713