Real time tire stiffness estimation using enhanced GDM and RLS for autonomous vehicles

Area of Science:

• Vehicle Dynamics and Control
• Machine Learning for Engineering Applications
• Automotive Safety Systems

Background:

• Yaw stability is critical for vehicle lateral control, heavily influenced by nonlinear tire-road interactions.
• Tire Lateral Stiffness (TLS) is a key parameter affecting yaw stability, varying with tire and road conditions.
• Precise TLS estimation is paramount for autonomous driving safety, particularly in demanding scenarios.

Purpose of the Study:

• To propose a novel framework for Tire Lateral Stiffness (TLS) identification.
• To explore modified Gradient Descent Methods (GDM) inspired by deep learning for TLS estimation.
• To develop and evaluate improved real-time algorithms for TLS tracking.

Main Methods:

• Established a theoretical link between Recursive Least Squares (RLS) and GDM, identifying RLS as a specific GDM case.
• Developed improved RLS variants for real-time TLS identification.
• Utilized simulations to compare various GDM and RLS algorithms under diverse conditions, including adaptive methods like Adam.

Main Results:

• Demonstrated effective TLS tracking capabilities across different algorithms and varying conditions.
• Adaptive GDM methods, such as Adam, showed superior performance in tracking TLS.
• Achieved a Relative Steady-State Error (RSSE) below 5% and a response time (t10) under 3 seconds with adaptive methods.

Conclusions:

• The proposed GDM-inspired framework effectively identifies TLS in real-time.
• Adaptive algorithms like Adam provide enhanced performance for TLS estimation in autonomous vehicles.
• Results offer practical insights for selecting appropriate estimators in safety-critical autonomous driving systems.