AI-Driven Anticipation of Vehicle Drift Dynamics

Anticipating vehicle drift is crucial for preventing critical driving scenarios, particularly in Autonomous Vehicles (AVs) and Advanced Driver Assistance Systems (ADAS). Integrating an intelligent system capable of predicting drift dynamics can significantly enhance both the safety and autonomy of modern vehicles. This paper proposes a data-driven AI methodology that combines Long Short-Term Memory (LSTM) and Multi-Layer Perceptron (MLP) neural network models to analyse vehicle kinematics and driver behaviour over time. By processing real-world data collected from professional drivers performing both regular and critical manoeuvres, the proposed system estimates the likelihood of entering a drift state, providing an early warning mechanism. The combined LSTM-MLP approach yields a probability score that serves as a threshold for proactive ADAS intervention, enabling timely corrective actions. The discussion also explores supervised and unsupervised learning approaches for segmenting driving states and enhancing drift detection accuracy. Additionally, it explores the impact of time-window selection in effectively capturing and anticipating the dynamics of this highly transient manoeuvre. The proposed approach and insights contribute to the advancement of AI-powered vehicle automation, reinforcing the role of predictive modelling in enhancing driving safety and stability.