Four-Wheel Independent Drive EVs: New Control Strategy Enhances Stability and Tracking Precision
In the rapidly evolving landscape of autonomous vehicle technology, one of the most critical challenges lies in ensuring both high-precision trajectory tracking and robust vehicle stability—especially during emergency maneuvers such as obstacle avoidance. As self-driving vehicles become increasingly common on public roads, their ability to respond safely and effectively under extreme conditions is paramount. A recent breakthrough from Tongji University has introduced a novel hierarchical control strategy that significantly improves the performance of four-wheel-independent-drive (4WID) electric vehicles in these demanding scenarios.
The research, led by Dr. Shuping Chen, Professor Zhiguo Zhao, and Kun Zhao from the School of Automotive Studies at Tongji University, presents an advanced coordinated control framework designed to simultaneously optimize path following accuracy and yaw stability. Published in the Journal of Tongji University (Natural Science), this work addresses a persistent challenge in autonomous driving: the inherent conflict between aggressive trajectory correction and vehicle dynamic stability.
Autonomous vehicles rely heavily on sophisticated control systems to interpret sensor data, plan motion paths, and execute driving commands. While many existing approaches focus primarily on minimizing lateral deviation from a reference path, they often overlook the risk of destabilizing the vehicle during sharp evasive actions. In high-speed or low-friction environments—such as wet or icy roads—sudden steering inputs can induce excessive sideslip angles or yaw rates, potentially leading to loss of control, skidding, or even rollover.
To address this issue, the Tongji team developed a two-layered control architecture that integrates linear time-varying model predictive control (LTV MPC) with quadratic programming-based torque distribution. This dual-layer approach allows for real-time optimization of both steering and propulsion forces, ensuring that the vehicle remains stable while accurately following its intended trajectory.
At the heart of the upper-layer controller is the LTV MPC algorithm, which continuously predicts the vehicle’s future behavior over a finite time horizon and computes optimal control actions. Unlike traditional MPC methods that use simplified single-track models, this implementation employs an 8-degree-of-freedom (DOF) vehicle model as the prediction engine. This enhanced model accounts for longitudinal, lateral, yaw, and roll dynamics, offering a more accurate representation of real-world vehicle behavior—particularly under combined acceleration and cornering conditions.
The MPC framework is further refined by embedding a PID-based speed tracking module directly into the optimization process. This integration ensures that longitudinal velocity deviations are actively minimized, allowing the controller to adapt to changing speed requirements without sacrificing lateral stability. By updating the predicted longitudinal velocity at each time step, the system maintains tighter control over overall vehicle motion, especially during acceleration or deceleration phases within complex maneuvers.
A key innovation in this study is the explicit inclusion of yaw stability as a core objective within the trajectory tracking framework. Instead of treating stability as a secondary concern, the controller generates both a desired front-wheel steering angle and an additional yaw moment to actively regulate the vehicle’s rotational dynamics. This additional yaw moment is not produced through conventional braking or steering adjustments but is instead achieved via differential torque delivery to the left and right wheels—a capability uniquely enabled by 4WID electric powertrains.
The lower layer of the control system is responsible for translating these high-level commands—total torque demand and desired yaw moment—into individual wheel torque allocations. Using a quadratic programming (QP) solver, the algorithm optimally distributes torque across all four wheels while considering multiple performance objectives: minimizing torque tracking error, reducing tire utilization, and limiting longitudinal slip energy loss.
Tire utilization, defined as the ratio of actual tire force to the maximum available friction force, serves as a critical indicator of stability margin. By keeping this value low, the control system preserves lateral force reserves, enabling the tires to respond effectively to unexpected disturbances. The QP formulation also incorporates physical constraints such as motor peak torque limits and road adhesion boundaries, ensuring that the solution remains feasible under real-world operating conditions.
To validate the effectiveness of their proposed method, the researchers conducted extensive simulations using a high-fidelity 14-DOF vehicle model equipped with a combined longitudinal-lateral brush tire model. This detailed plant model captures complex interactions between suspension dynamics, tire deformation, and load transfer, providing a realistic testbed for evaluating control performance under extreme conditions.
The primary test scenario was the double lane-change maneuver—a standard benchmark for emergency avoidance performance. Simulations were performed under varying speeds (36 km/h, 72 km/h, and 90 km/h), different road friction coefficients (μ = 0.85, 0.8, and 0.3), and with and without active stability control enabled. The results demonstrated consistent improvements in both tracking accuracy and dynamic stability across all test cases.
At higher speeds, where the risk of instability increases significantly, the controlled vehicle maintained lateral deviations below 0.28 meters—well within acceptable safety margins. Even under low-adhesion conditions (μ = 0.3), the system prevented excessive sideslip angles and yaw rates, keeping the vehicle within its handling envelope. Notably, when compared to a baseline controller that only focused on trajectory tracking, the integrated stability-aware approach reduced lateral tracking error by up to 35% and improved heading angle consistency by over 40%.
One of the most compelling findings was the system’s ability to reduce driver workload—or in the case of autonomous vehicles, computational burden—by minimizing the required steering input. By generating corrective yaw moments through torque vectoring rather than relying solely on front-wheel steering, the controller allowed for smaller steering angles, which in turn reduced tire stress and improved ride comfort. This effect was particularly pronounced on slippery surfaces, where large steering inputs could easily exceed the tire’s friction limits.
Moreover, the use of LTV MPC provided superior adaptability compared to fixed-gain controllers. Because the system linearizes the vehicle dynamics around the current operating point at each time step, it can account for changes in speed, load distribution, and road conditions in real time. This time-varying nature makes the controller inherently more robust than linear time-invariant (LTI) alternatives, which assume constant system parameters.
The brush tire model used in the simulation also contributed to the realism of the results. Unlike simplified linear tire models that are only valid within small slip ranges, the brush model captures the nonlinear coupling between longitudinal and lateral forces, especially near the friction limit. This fidelity is essential for accurately predicting vehicle behavior during aggressive maneuvers, where tires often operate in the saturation region.
From a practical standpoint, the proposed control strategy is well-suited for integration into modern electric vehicle platforms. Most 4WID EVs already possess the necessary hardware—individual motor controllers, high-speed communication networks, and advanced sensors—to support such a system. The computational demands of LTV MPC and QP-based allocation are manageable with current automotive-grade processors, especially given the relatively short prediction horizons typically used in real-time control applications.
Another advantage of this approach is its modularity. The separation between the upper-level motion planner and the lower-level actuator allocator allows for flexible tuning and adaptation. For instance, different weighting matrices in the MPC cost function can be used to prioritize comfort, efficiency, or sportiness depending on driving mode. Similarly, the QP objective can be adjusted to emphasize energy conservation in eco-mode or maximum grip in performance mode.
The implications of this research extend beyond academic interest. As regulatory bodies and safety organizations like Euro NCAP and IIHS place increasing emphasis on automated emergency avoidance capabilities, control strategies that enhance both safety and precision will become essential for market competitiveness. Automakers investing in autonomous driving technologies can leverage such innovations to meet stringent safety standards while delivering a smoother, more confident driving experience.
Furthermore, the methodology presented here lays the groundwork for future advancements in integrated vehicle dynamics control. The authors suggest that future work could explore the inclusion of roll dynamics and rollover prevention metrics, particularly for taller vehicles such as SUVs and electric vans. Additionally, adaptive tuning of the QP weighting coefficients—using machine learning or observer-based estimation—could further enhance performance across diverse driving conditions.
In conclusion, the research conducted by Chen, Zhao, and Zhao represents a significant step forward in the field of autonomous vehicle control. By unifying trajectory tracking and yaw stability within a single, computationally efficient framework, their approach offers a practical solution to one of the most pressing challenges in modern mobility. As 4WID electric vehicles continue to gain traction in the automotive market, control strategies like this one will play a crucial role in shaping the future of safe, intelligent transportation.
The successful implementation of such systems not only enhances safety but also builds public trust in autonomous technologies. When vehicles can navigate emergency situations with the precision and composure of an expert human driver—or better—consumer confidence in self-driving capabilities will grow. This, in turn, accelerates the adoption of electrified and automated mobility solutions, contributing to broader goals of sustainability, traffic efficiency, and accident reduction.
As urban environments become more congested and road safety remains a global priority, innovations in vehicle dynamics control will remain at the forefront of automotive engineering. The work from Tongji University exemplifies how academic research can directly inform and improve real-world technologies, bridging the gap between theoretical control theory and practical application.
With continued development and validation—both in simulation and on test tracks—this control strategy could soon find its way into production vehicles. It stands as a testament to the power of interdisciplinary research, combining elements of control theory, mechanical engineering, and artificial intelligence to solve complex, real-world problems.
Ultimately, the goal of autonomous driving is not just to replicate human driving but to surpass it in terms of safety, efficiency, and reliability. The coordinated control method developed by the Tongji team brings us one step closer to that vision, demonstrating that with the right algorithms and system architecture, electric vehicles can handle even the most challenging driving scenarios with grace and precision.
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