New AI physics system enhances EV safety by predicting loss of control in real-time. Learn how this tech is revolutionizing autonomous driving! Read more.
As the automotive industry accelerates towards electrification and automation, software is becoming increasingly central to ensuring vehicle safety and efficiency. Modern electric vehicles (EVs) are no longer simply mechanical machines; they are complex systems where algorithms constantly assess motion to make split-second decisions.
The Challenge of Accurate Vehicle State Estimation
One of the biggest challenges for self-driving cars is accurately understanding how the vehicle is moving on the road. Even a small error in estimating speed, acceleration, or slip angle can lead to delayed braking or steering responses. In automated driving environments, these errors can quickly accumulate, increasing the risk of losing control.
Therefore, engineers view vehicle state estimation as the cornerstone of future transportation. However, traditional physics-based models, built on ideal assumptions, are showing clear limitations. Real-world roads are constantly changing, tires deform, grip varies, and sudden steering maneuvers occur frequently, making it difficult for classical models to keep up.
A Hybrid Approach: AI Physics
In response, a research team led by Professor Kanghyun Nam at DGIST has proposed a new approach. Instead of relying solely on physics or completely deferring to artificial intelligence (AI), the team combined both in an AI physics-based state estimation system. This project, a collaboration with Shanghai Jiao Tong University and the University of Tokyo, highlights the interdisciplinary and international nature of the research.
Predicting Slip Angle for Enhanced Stability
The system focuses on estimating vehicle states that onboard sensors cannot directly measure, most importantly the slip angle. This parameter reflects the degree to which a vehicle slides sideways when cornering or moving on slippery surfaces. If not detected in time, a significant slip angle can cause the vehicle to become unstable before the control system can react.
The problem lies in the fact that tire behavior is not fixed. It changes with speed, road surface, and operating conditions, making traditional estimation methods less effective. To address this, the research team built a hybrid estimation framework where a physics-based tire model is augmented by the learning capabilities of AI.
Combining Physics and AI for Robust Performance
Specifically, the system combines a physical tire model with an AI-based regression method. Data from sensors measuring tire lateral force is continuously fed in, allowing the model to adapt to nonlinear behaviors and environmental changes. At the core of this architecture is a scent-free Kalman filter, integrated with Gaussian process regression.
The Kalman filter ensures physical consistency, preventing the system from making unrealistic estimates. Meanwhile, the AI component provides flexibility, allowing the model to learn from data and compensate for discrepancies that pure physics cannot explain. This combination creates a system that is both accurate and reliable.
Real-World Testing and Validation
To verify its effectiveness, the researchers tested the system on a real electric vehicle platform. The tests included various road surfaces, speeds, and diverse cornering scenarios. The results showed that the system maintained high accuracy in all conditions, a key factor for real-world application.
Benefits for Vehicle Safety and Efficiency
Accurate vehicle state estimation benefits many critical functions, from stability control and ensuring the safety of self-driving cars to optimizing energy efficiency. When the system detects early signs of instability, it can intervene more promptly and accurately, minimizing risks for vehicle occupants.
Professor Kanghyun Nam notes that the research team focused not only on accuracy but also on long-term reliability. He believes that combining a physics model and AI has filled the gaps left by traditional methods, opening up new avenues for vehicle control architectures in the future.
A Step Towards Intelligent Vehicles
Published in IEEE Transactions on Industrial Electronics, the research marks a significant step in the effort to integrate AI into physical control without compromising safety. As electric and self-driving vehicles become increasingly prevalent, systems like this could become the foundation for the next generation of intelligent vehicles.