The enigmatic curvature of a curling rock’s trajectory has long captivated scientists and athletes alike. Thispaper presents a novel approach to understanding the curlingphenomenon, leveraging machine learning-enhanced moleculardynamics (ML-MD) simulations to investigate the quasi-liquidlayer (QLL) and its pressure-dependent asymmetry. We introducea new ML-MD force field trained on high-fidelity ab initio data,enabling accurate simulation of ice surface dynamics under theextreme pressures exerted by a curling rock. Our simulationsreveal a significant pressure asymmetry in the QLL thicknessand viscosity, directly correlating with the observed curl. Furthermore, we propose a refined theoretical model incorporating thesepressure-dependent QLL properties, providing a more accurateprediction of the curl distance. This research offers a significantadvancement in our understanding of the curling phenomenon,bridging the gap between microscopic dynamics and macroscopicbehavior