As a branch of reinforcement learning, genetic algorithm has shown significant advantages on exploring the multi-dimensional space in artificial neural networks. Although there are many great applications online applying genetic algorithms to neural networks, 2D platform game AI agents are rarely seen. This project tries to create self-learning competitive AI agents in a 2D platform 2-player shooting game using artificial neural network and genetic algorithm. The game provides three modes to allow human players to either play the game freely or train their own AI agents to fight against. Unlike traditional game AI in this type of game, my AI agent makes real-time decisions based on the input of the human player, no predefined strategies were written into the agents.