Dynamic Uncertainty-Adaptive Vehicle Lateral Velocity Estimation for Near-Limit Cornering With Tire Stiffness Compensation

Electric vehicles (EVs) face significant challenges in lateral velocity estimation during high-speed cornering maneuvers. Traditional estimation methods often fail due to tire model uncertainties and measurement biases under intense external excitation. To address these limitations, we propose an uncertainty-adaptive unscented Kalman filter (UA-UKF) framework that incorporates virtual measurements using readily available chassis signals. Our approach features an adaptive tire stiffness model that effectively captures nonlinear tire behavior near the handling limits. We introduce a dynamic confidence scaling mechanism that adjusts the tire stiffness estimates based on real-time model uncertainty and excitation levels. The framework is enhanced by neural network-generated virtual measurements that compensate for inertial measurement biases caused by vehicle body roll. Extensive validation on the Dongfeng E70 platform demonstrates superior performance across diverse driving scenarios, including high-speed maneuvers on both asphalt and ice-snow roads. In circular acceleration tests, our method reduces the mean and maximum lateral velocity estimation errors by 41.1% and 38.0%, respectively, compared to conventional UKF.