Neural Approaches for Vehicle Dynamics: Side Slip Angle Estimation and Traction Control

This thesis investigates neural-based learning methods for vehicle dynamics estimation and control, with emphasis on side slip angle estimation and traction control in high-performance vehicles. Traditional control strategies, based on analytical models and offline calibration, are often limited by uncertainties, sensor noise, and rapidly changing conditions. Data-driven approaches using Deep Learning (DL), Reinforcement Learning (RL), and Neural Model Predictive Control (NMPC) offer greater adaptability, robustness, and accuracy in these scenarios. The first part focuses on estimating the lateral dynamics of a racing vehicle using neural architectures such as Feedforward, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) networks. These models were trained and validated through simulations using telemetry data from Podium Advanced Technologies, demonstrating their capability to function as reliable virtual sensors for state estimation. The second part presents a reinforcement learning–based traction control system. Different agents, including Deep Deterministic Policy Gradient (DDPG), Twin Delayed DDPG (TD3), Soft Actor–Critic (SAC), and Proximal Policy Optimization (PPO), were implemented in MATLAB/Simulink to learn optimal control policies for maintaining grip and stability. SAC and PPO achieved the most stable and consistent results under variable driving conditions. Finally, a Neural Model Predictive Control approach was explored as an alternative to RL-based control. By replacing the analytical vehicle model with a trained neural predictor, NMPC preserves the optimization framework of traditional MPC while improving flexibility in modeling nonlinear dynamics. The results confirm that neural and reinforcement learning methods can significantly enhance both estimation and control in modern vehicles. Their data-driven nature and ability to model complex dynamics represent a promising foundation for future traction control systems and advanced driver-assistance technologies.