Research of intelligent air combat model based on reinforcement learning

Abstract

The development in artificial intelligence has dramatically changed all industries, among which AI-assisted air combat is a representative case of success. An Intelligent air combat model that consists of the attainment of samples and a decision-making model is constructed in connection with air combat simulator. Considering the characteristics of air combat continuous states and actions, DQN algorithm is selected as the model of intelligent air combat by comparison among several algorithms. Meanwhile, the AI network is trained interactively with AI enemies in the air combat simulation game DCS World, resulting in a model that is able to manipulate aircraft to a degree and many cases of air combat, by analyzing which a collection of winning, losing and dual samples is derived. The result of simulation indicates that the Intelligent air combat model has certain ability to generate strategic samples and enrich tactics in air combat environments.

Keywords

• air combat|deep reinforcement learning|flight simulation game