A self-built reinforcement learning system integrating Unreal Engine 5.5 (C++), PyTorch + CUDA, and a hybrid-action Proximal Policy Optimization (PPO) algorithm capable of training a virtual ground-to-air agent to intercept dynamic targets in a 3D simulated physics-driven environment.