PGCLN: A physics-guided approach for personalized optimal sitting posture prediction using pressure distribution

To bridge the gap between physiological data and engineering implementation, this study addressed the limitation of indirect mapping of sitting posture by establishing a direct predictive link between physiological information from pressure distribution and optimal sitting posture. A novel predicting framework named Physics-Guided and Contrastive Learning Network (PGCLN) was proposed to predict sitting posture using 12 pressure distribution features and seven anthropometric data, and designed to overcome small-sample and high-dimensional regression problems. Crucially, Contact Pressure Distribution Index (CPDI) was firstly introduced as a new biomechanical constraint to ensure the model’s physical consistency. By integrating CPDI-based physical prior constraint with contrastive learning, the PGCLN can accurately predict the optimal sitting posture for diverse occupants with a R 2 of 0.9711 and RMSE of 0.7649, significantly outperforming baseline models. Furthermore, SHAP analysis reveals that Waist Average Pressure and Shoulder and Back Contact Area are the primary determinants for PGCLN. Overall, this work offers a robust solution for an on-vehicle deployment of real-time, personalized seat adaptive adjustment, providing both a theoretical foundation and a practical pathway for intelligent seating systems in human factors engineering.