Researchers have developed a new physical artificial intelligence system that enables electric vehicles to detect loss of control in real time by combining traditional physics-based modeling with AI, according to newly published research. The system is designed to improve vehicle stability detection and could play a key role in future EV safety and autonomous driving systems.
The research, led by Professor Kanghyun Nam at DGIST in collaboration with Shanghai Jiao Tong University and the University of Tokyo, focuses on improving how vehicles estimate motion states that cannot be directly measured by onboard sensors. One of the most critical of these is the sideslip angle, which indicates how much a vehicle slides sideways during cornering or low-friction conditions.
As the researchers explain, “Through a new approach that combines physical models and AI, we can estimate the driving conditions of electric vehicles with greater precision and reliability.”
Conventional vehicle control systems rely heavily on mathematical models that assume predictable tire and road behavior, but those assumptions often break down in real-world driving.
Professor Nam highlighted the long-term potential of the work. He said the team focused on improving reliability as much as precision. Nam emphasized that combining physics and artificial intelligence helped close gaps left by traditional models.
The new physical AI approach blends a tire force physics model with AI-based regression techniques, allowing the system to adapt continuously as conditions change. The researchers noted that “existing vehicle dynamics models struggle when tire behavior becomes highly nonlinear,” a limitation the new system is designed to overcome.
The system processes lateral tire force sensor data to estimate motion states in real time, including scenarios involving different road surfaces, speeds, and aggressive cornering.
During testing, the researchers reported that the physical AI estimator “maintained stable accuracy even under conditions where conventional estimators showed large deviations.” This capability is especially important for EVs, where instant torque delivery can increase the risk of traction loss.
Physical AI refers to AI systems that learn directly from interactions with the physical world rather than relying solely on abstract data. In automotive applications, this allows AI models to respect real-world constraints such as friction limits, inertia, and force balance. According to the research team, “integrating physics knowledge into AI improves robustness and reduces unpredictable behavior,” particularly in safety-critical systems.
As EVs move toward higher levels of automation, accurate motion estimation becomes increasingly essential. The study states that “real-time detection of loss of control can significantly enhance vehicle safety and stability control performance.”
Researchers believe manufacturers could integrate the system into future EV platforms as part of advanced driver assistance or autonomous driving stacks. Rather than replacing existing control systems, the physical AI model would augment them with more reliable motion awareness, especially in edge cases where conventional estimators fail.