Despite significant advancements in applying artificial intelligence to Chinese chess, challenges such as inefficient training, limited system autonomy, and poor integration of perception, strategy, and execution persist. This study presents a fully integrated intelligent Chinese chess system that unifies visual perception, reinforcement learning-based strategy, and robotic control. By leveraging a JAX-based Pgx platform for efficient parallel training, an AlphaZero-style algorithm with Monte Carlo Tree Search for strategic decision-making, and an optimized SCARA robotic arm for precise move execution, the system achieves a piece recognition accuracy of 99.39% and generates rule-compliant moves with emergent defensive capabilities. Experimental results demonstrate robust performance across all modules, establishing a foundation for advancing intelligent gaming robots in human-robot interaction, education, and entertainment.