Electric vehicles increasingly rely on software to stay stable, efficient, and safe.

As cars take on more automated driving tasks, they must interpret complex road conditions faster than human drivers.

That challenge has pushed engineers to rethink how vehicles understand their own motion.

Modern control systems depend on precise knowledge of how a vehicle moves at any moment.

Even small errors can affect braking, steering, and stability. In autonomous systems, those errors can compound quickly.

Engineers, therefore, view vehicle state estimation as one of the most critical foundations of future mobility.

Researchers now argue that traditional modeling approaches alone cannot keep up. Real roads introduce unpredictable factors such as tire deformation, surface changes, and sudden maneuvers.

These effects often fall outside the assumptions built into classical vehicle models.

Understanding vehicle behavior

A research team led by Professor Kanghyun Nam at DGIST has developed a new solution to that problem.

The group created a physical AI-based vehicle state estimation system designed to track how electric vehicles behave in real time.

The project involved international collaboration with Shanghai Jiao Tong University and the University of Tokyo.

The system focuses on estimating vehicle motion states that sensors cannot measure directly.

One of the most important of these is the sideslip angle. This value indicates how much a vehicle slides sideways during turns or low-friction conditions.

Sideslip plays a major role in vehicle stability. When drivers or automated systems fail to detect it early, control systems may react too late.

Conventional estimation methods struggle because tire behavior changes constantly.

Road surfaces and speed further complicate the calculations.

To overcome these limitations, the research team designed a hybrid estimation framework.

The approach combines physical vehicle models with artificial intelligence. Instead of replacing physics, the system strengthens it with data-driven learning.

The framework merges a physical tire model with an AI-based regression method.

Sensor data measuring lateral tire force feeds into the system continuously. This allows the model to adapt to nonlinear tire behavior and environmental variation.

At the core of the system is an unscented Kalman filter observer integrated with Gaussian process regression.

The Kalman filter ensures physical consistency. The AI component adds flexibility and learning capability.

Together, they enable faster and more accurate estimation than traditional approaches.

Testing and implications

The researchers validated the system using an actual electric vehicle platform.

Tests covered multiple road surfaces, speeds, and cornering scenarios. The system maintained strong accuracy across all conditions.

Engineers see this consistency as essential for deployment in real vehicles.

Accurate vehicle state estimation supports several critical functions.

These include stability control, autonomous driving safety, and energy efficiency. Better estimates allow control systems to intervene earlier and more precisely.

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.

He said, “Through a new approach that combines physical models and AI, we can estimate the driving conditions of electric vehicles with greater precision and reliability.”

Researchers believe the approach could shape future vehicle control architectures.

The system offers a path toward AI-assisted physical control without sacrificing reliability.