Automobile Lateral Control Based on Driving Style Classification Using AM-LSTM Method

Many driver-assistance systems adopt unified control strategies that overlook inter-driver heterogeneity, thereby limiting the performance of shared control. This study develops a driver-specific model to support personalized camber alignment control by predicting the driver’s next-step desired lateral force and mapping it to front and rear feedforward camber gains. The approach applies FA to extract compact and interpretable latent features from driver and vehicle signals, and employs a GMM to partition short-horizon driving behavior into calm, moderate, and aggressive styles, enabling probabilistic gating. A style-conditioned AM-LSTM is then used to regress the camber gains based on the identified driving style. The model is trained and evaluated using a human-in-the-loop simulator dataset. Experimental results show that the GMM provides stable style segmentation and effective gating, while FA enhances GMM classification and improves overall model accuracy. These findings demonstrate that the proposed GMM-LSTM framework offers an accurate and interpretable driver model suitable for integration with personalized active chassis control, aiming to improve driver performance under manageable workload.