<p>Selective laser melting (SLM) demands precise coordination between scanning motion and laser power modulation to achieve stable melt pool behavior and high geometric accuracy. However, mismatches in dynamic responses between scanning and power systems often degrade synchronization, especially under high-speed motion and sharp trajectory transitions. To address this issue, this paper proposes a master-slave cross-coupled iterative learning control (CCILC) framework to improve scan-power coordination in SLM processes with robust convergence under system uncertainties analytically established. In the proposed approach, the scanning system is designated as the master and the power modulation system as the slave, allowing laser power commands to adapt to actual scanning behavior. CCILC is employed to jointly reduce tracking and contour errors, while disturbance observers and robust feedback controllers are incorporated to enhance robustness against system uncertainties. In addition, a time-varying zero-phase learning filter is introduced to adapt the learning bandwidth according to error magnitude, effectively suppressing high-frequency oscillations in corner regions while maintaining stable convergence. Experimental results obtained from an EtherCAT-based SLM platform demonstrate that the proposed method achieves reductions of 31.65% in scanning tracking errors, 22.12% in contour errors, and 90.07% in power tracking errors compared with conventional distributed control. In addition, the proposed method further reduces residual errors in the corner regions, where the previous hybrid control method still exhibited noticeable drift and local error concentration. The error analysis further shows clearer improvement in these difficult-to-compensate regions, indicating that the proposed framework effectively improves scan-power coordination and tracking robustness in challenging trajectory sections.</p>