New physical AI lets EVs detect loss of control in real time

Researchers led by Professor Kanghyun Nam at DGIST, in collaboration with Shanghai Jiao Tong University and the University of Tokyo, have developed a novel physical AI-based system to improve real-time vehicle state estimation for electric vehicles (EVs). This system addresses the critical challenge of accurately detecting vehicle motion states—especially the sideslip angle, which indicates sideways sliding during turns or low-friction conditions and is vital for vehicle stability. Traditional models struggle with unpredictable real-world factors like tire deformation and varying road surfaces, but the new hybrid framework combines physical tire models with AI-driven regression to adapt dynamically to these nonlinear behaviors. At the core of the system is an unscented Kalman filter observer integrated with Gaussian process regression, which together ensure both physical consistency and learning flexibility. Tested on an actual EV platform across diverse road conditions and speeds, the system demonstrated strong accuracy and robustness. This advancement enables earlier and more precise interventions by stability control and autonomous driving systems, enhancing safety and energy efficiency. Professor Nam emphasized that