Method for training neural network based on genetic algorithm to control game role behaviors

Abstract

The invention discloses a method for training a neural network to control game role behaviors based on a genetic algorithm. According to the method, necessary input information is obtained from each frame in a game engine through population samples; behavior input information is obtained through a neural network; the samples and the environment are interacted to update a fitness function, after acertain time, the population samples are divided into two sample populations according to a certain proportion and multiplied in two modes respectively, progenies of the two populations are combined into one population, and the population more adapting to the environment is obtained along with ceaseless training. According to the method, the neural network is used for processing environment input;more comprehensive information can be obtained; compared with the prior art, the method is more flexible in action without behavior pattern limitation, action strategy optimization can be automatically carried out according to actions of players in the actual application process, the genetic algorithm used in the method enables the search range to be wider, the sample diversity is increased, andthe used multi-genetic strategy makes up for the defect that local optimum is likely to enter too early to a certain extent.

Description

technical field [0001] The invention belongs to the field of game character control, and in particular relates to a method for training a neural network based on a genetic algorithm to control the behavior of a game character. Background technique [0002] With the development of game 3D technology and physical simulation technology, the future game industry that can be seen will gradually occupy people's leisure time. In recent years, some game masterpieces have also slowly entered the market, and whether the behavior of autonomous characters in the game has the ability to change behavior patterns has always been a concern of players. [0003] At present, most game AIs are mostly implemented through finite state machines. The realization of this method requires designers to design a complete set of behavioral logic, and at the same time, various transition conditions need to be considered. The advantage lies in stability, and the disadvantages are also obvious. The behavio...

AI Technical Summary

Problems solved by technology

The advantage lies in stability, and the disadvantages are also obvious. The behavior is predictable, and there will be situations that human beings cannot consider, which will lead to BUG.

In this case, not only the interest of players will decrease rapidly, but the word-of-mouth of the game company will also be affected

Images