New Adaptive Algorithm Enhances Electric Vehicle State Estimation Accuracy
In the rapidly evolving world of intelligent transportation, the ability to accurately estimate a vehicle’s dynamic state in real time has become a cornerstone of advanced driver assistance systems (ADAS) and autonomous driving technologies. As vehicles grow smarter and more connected, the demand for precise, reliable, and robust state estimation—particularly under extreme driving conditions—has intensified. A recent breakthrough from researchers at Shijiazhuang Tiedao University has introduced a novel algorithm that significantly improves the accuracy and reliability of vehicle state estimation by addressing a critical yet often overlooked factor: the time-varying nature of tire cornering stiffness.
Published in the January 2024 issue of Mechanical Science and Technology for Aerospace Engineering, the study led by Fu Yuesheng, a master’s student, and co-authored by Professor Li Shaohua and Wang Guiyang from the State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures, presents a groundbreaking approach to vehicle state estimation. The research introduces an embedded algorithm known as FAEKF+FFRLS, which combines fuzzy adaptive extended Kalman filtering with a forgetting-factor recursive least squares method to enable real-time updates of tire cornering stiffness—a parameter that directly influences vehicle handling, stability, and safety.
The significance of this development lies in its departure from conventional modeling assumptions. Most existing vehicle state estimation models rely on linear tire models, which assume that tire cornering stiffness remains constant regardless of driving conditions. However, in reality, tire behavior is highly nonlinear, especially during aggressive maneuvers, low-friction scenarios, or high-speed cornering. Under such conditions, the assumption of constant stiffness leads to significant estimation errors, undermining the performance of control systems that depend on accurate state data.
“Traditional methods often neglect the time-varying characteristics of tire parameters,” explains Professor Li Shaohua, the corresponding author of the paper. “This oversight becomes particularly problematic when estimating vehicle states under limit conditions, where tire forces operate in their nonlinear region. Our goal was to develop an algorithm that not only adapts to changing noise environments but also dynamically updates key model parameters to reflect real-world physics.”
The FAEKF+FFRLS algorithm achieves this through a dual-layer estimation framework. At its core, the method integrates two powerful estimation techniques into a unified, embedded structure. The first component, the Fuzzy Adaptive Extended Kalman Filter (FAEKF), is responsible for estimating the vehicle’s longitudinal speed, yaw rate, and sideslip angle—three fundamental states that define a vehicle’s motion behavior. Unlike the standard Extended Kalman Filter (EKF), which uses fixed noise covariance matrices, the FAEKF employs a fuzzy logic controller to dynamically adjust the Kalman gain based on the discrepancy between predicted and actual measurement residuals.
This adaptive mechanism allows the filter to respond intelligently to changes in sensor noise, which can vary due to environmental factors, sensor degradation, or signal interference. By continuously tuning the filter’s sensitivity, the FAEKF maintains high estimation accuracy even when measurement quality fluctuates—a common challenge in real-world driving scenarios.
The second component of the algorithm, the Forgetting-Factor Recursive Least Squares (FFRLS) method, focuses on estimating the tire cornering stiffness in real time. Rather than treating stiffness as a static parameter derived from offline calibration, the FFRLS model treats it as a dynamic variable that evolves with driving conditions. It does so by analyzing the relationship between lateral acceleration, yaw rate, and steering input, using a side-slip and yaw coupling formulation that avoids direct dependence on the sideslip angle—an inherently difficult parameter to measure or estimate accurately.
What sets this approach apart is the seamless integration between the two components. Instead of running independently, the FAEKF and FFRLS modules are embedded within a single computational loop. The FAEKF first predicts the vehicle’s state, which is then fed into the FFRLS module to estimate the current tire cornering stiffness. This updated stiffness value is subsequently used to refine the vehicle dynamics model within the FAEKF, leading to more accurate state predictions in the next iteration. This closed-loop interaction enables mutual correction between state and parameter estimation, enhancing both precision and robustness.
To validate the effectiveness of the proposed algorithm, the research team conducted extensive co-simulations using Trucksim and MATLAB/Simulink. They tested the algorithm under three distinct driving scenarios: low-speed on a low-friction surface (30 km/h, μ = 0.3), high-speed on a high-friction surface (80 km/h, μ = 0.8), and high-speed on a low-friction surface (80 km/h, μ = 0.3). These conditions were chosen to represent a wide range of operational extremes, from normal urban driving to aggressive maneuvers on slippery roads.
The results were compelling. Across all test cases, the FAEKF+FFRLS algorithm demonstrated superior performance compared to the standard EKF. In particular, the estimation error for yaw rate and sideslip angle was significantly reduced, especially in high-speed, low-grip conditions where tire nonlinearity is most pronounced. For instance, in the most challenging scenario—80 km/h on a μ = 0.3 surface—the relative error in sideslip angle estimation was reduced from 41.0% with the standard EKF to just 5.36% with the proposed method. Similarly, yaw rate estimation error dropped from 10.09% to 3.88%, showcasing the algorithm’s ability to maintain accuracy even under severe dynamic loading.
Moreover, the algorithm exhibited excellent stability and robustness. Unlike some adaptive filters that may diverge under rapid changes in noise or model uncertainty, the FAEKF+FFRLS maintained consistent performance throughout the simulation runs. The fuzzy controller effectively prevented over-correction, while the forgetting factor in the FFRLS ensured that outdated data did not unduly influence the stiffness estimate, allowing the model to adapt quickly to new conditions.
One of the key advantages of this approach is its practicality for real-world implementation. The computational complexity of the algorithm remains manageable, making it suitable for deployment on embedded automotive control units. Additionally, the method leverages data that is readily available in modern electric vehicles, particularly those with in-wheel motors. By utilizing motor torque and rotational speed measurements, the researchers were able to compute tire longitudinal forces directly, eliminating the need for additional sensors and improving the fidelity of the vehicle dynamics model.
This focus on distributed electric vehicles is particularly timely. As the automotive industry shifts toward electrification and advanced drivetrain architectures, the availability of high-resolution actuator data opens new possibilities for model-based estimation. The work by Fu, Li, and Wang exemplifies how these technological advancements can be harnessed to improve fundamental vehicle functions such as state estimation.
Beyond its immediate application in ADAS and autonomous driving, the algorithm has broader implications for vehicle safety and control. Accurate estimation of sideslip angle, for example, is critical for electronic stability control (ESC) systems, which rely on this parameter to detect and correct loss of traction. By providing a more reliable estimate of this elusive state, the FAEKF+FFRLS algorithm could enhance the performance of ESC, particularly in edge cases where traditional systems may struggle.
Similarly, the real-time estimation of tire cornering stiffness could be used to infer road surface conditions, enabling predictive control strategies that adjust vehicle behavior based on anticipated grip levels. This capability aligns with the growing trend toward proactive safety systems that anticipate hazards rather than simply reacting to them.
The research also highlights the importance of interdisciplinary collaboration in advancing automotive technology. The algorithm draws upon principles from control theory, signal processing, fuzzy logic, and vehicle dynamics, demonstrating how the integration of diverse methodologies can yield solutions that outperform traditional, single-discipline approaches.
Looking ahead, the authors suggest several directions for future work. One area of interest is the extension of the algorithm to include additional vehicle parameters, such as road friction coefficient or vehicle mass, which also exhibit time-varying behavior. Another potential enhancement involves incorporating machine learning techniques to further refine the adaptation mechanism, possibly enabling the system to learn from past driving experiences and improve over time.
Additionally, the team plans to conduct real-world testing using instrumented vehicles to validate the algorithm under actual driving conditions. While simulation results are promising, field testing will be essential to assess the algorithm’s performance in the presence of unmodeled dynamics, sensor noise, and unpredictable driver behavior.
The publication of this research in Mechanical Science and Technology for Aerospace Engineering underscores its relevance not only to the automotive sector but also to broader domains involving dynamic system estimation. The rigorous methodology, comprehensive validation, and clear presentation of results reflect a high standard of scientific inquiry.
In an era where vehicle safety and autonomy are paramount, the work of Fu Yuesheng, Li Shaohua, and Wang Guiyang represents a significant step forward in the quest for more intelligent, responsive, and trustworthy vehicle systems. By acknowledging and addressing the dynamic nature of tire behavior, their algorithm sets a new benchmark for state estimation in intelligent electric vehicles.
As the transportation landscape continues to evolve, innovations like the FAEKF+FFRLS algorithm will play a crucial role in shaping the future of mobility—making it safer, smarter, and more adaptive to the complexities of real-world driving.
Fu Yuesheng, Li Shaohua, Wang Guiyang, Mechanical Science and Technology for Aerospace Engineering, DOI: 10.13433/j.cnki.1003-8728.20220190