DRL-LANCP (Link: IEEE TMC) investigates how to improve the generalization of low-altitude radio map prediction by disentangling propagation-related representations, bridging data learning with implicit physical structure.