The Application Of Neural Network In Real-Time Strategy Game

In today’s computer games, the development of graphics quality has reached a level of near-extreme, artificial intelligence has become a significant factor which can decide whether it’s a successful game or not. The proportion of effort and money game developers invest in artificial intelligence is also growing.So far, basically the game artificial intelligence are finite state machines, scripting languages, genetic algorithm, neural networks, etc. Today, however, the domestic game artificial intelligence is mostly state machine. The state machine is too simple, not smart enough. once people know the rules,it is very easy to crack. Besides, the achievement of state machine needs too many if-else statements. For this situation, this paper proposes applying neural network in the game and using the backpropagation algorithm for supervised training.First of all this paper researches and analysises of the principles of finite-state machine, neural network and back-propagation algorithm. Neural network matches input among learning samples, and returns output which matches the most, this shows neural network is more intelligent than state machine. Meanwhile, neural network does not need to write too many if-else statements, because the information which stat machine needs stores in trained neural network’s weights. Back propagation algorithm is the essence of the gradient descent algorithm,using the error value reduction to make the actual output matches the expected output.Based on the analysis of real-time strategy game rules, trains the sample data, gives an overall design of game AI based on neural network; designs a series of specific logic of the game in detail:push the line, home, in pursuit of the enemy hero, Etc.; and gives a concrete realization of the solution. This paper makes use of organic combination of neural network and finite state machine, succeeds in applying neural network into the real-time strategy games. This paper applies neural network to make decisions for the computer player, while executing the decision by using state machine.Finally,this paper has carried on the test to the implemented AI game software. This software since its launch on Tencent QQ game, after inspected by lots of game players, widely acclaimed, it has the certain difficulty for the novice game player.