Reaserch On The Decision-making Method Of Non-player Roles In Real-Time Against Games

The non-player roles in the game(game AI)have been the research hotspots of game artificial intelligence and simulation platform intelligence.The intelligence of game AI plays an important role in improving the gaming experience for game players.With the rapid development of the game industry,game AI's decision-making research is of great significance in the game field.The simple machine learning method often fails to realize the role behavior's autonomous behavior in complex scenarios.Given the environmental parameters of the instant games and the stock forecasting data in quantitative finance domain both have independent identical distribution features,based on the deep study about LSTM(Long-Short Term Memory)model of stock prediction has got very good result,therefore,after the further study between the correlation of the decision-making game environment parameters and the game AI,this paper designed the two-level decision-making mechanism:in against decision-making,game AI must first get strategy categories TAB,then select the tactical decisions on the basis of strategic decision category label.According to two different demand for independent game AI and united AI scenarios,this paper designed different decision making method based on the deep learning.The main work of this paper can be summarized as follows:Analysis and collection of decision sample attributes:analyzing the correlation between game environment parameter attributes and independent game AI,also about the multiple game AI,this paper designed the decision-making samples's characteristic attribute for two demand of the game environmental;Using the Shared memory method of Windows API,simulate the behavior of human players and collect the environment parameter sample data set based on two requirements of game AI.Independent decision model for the individual game AI:the first level strategic decision contains only two categories TAB:attack and retreat,so this paper is designed the fast LSTM individual game AI strategic decision model.For secondary category label has more tactical decisions and the few samples,to avoid over fitting,this paper using the BLSTM(Bidirectional-LSTM)model framework which can make full use of the whole sample sequence information.Both two-stage decision-making models improve the decision accuracy of the model by training the number of layers and parameters.Experimental results show that the hierarchical of two level decision methods based on independent AI has merits of high precision,fast speed and practical value.In addition,based on the requirements of game AI path planning,the JPS+ pathfinder algorithm is used to realize more effective route planning.Multiple game AI's united strategic decisions:focus on the game mode of multi-game AI collaborative operation,this paper proposes two schemes for multi-game:United AI mechanism and United AI model.In united AI mechanism,multiple game AI have made independent strategic decisions and in order to comply with the team's benefit maximization as the guidance basis of team collaborative strategy,coordinated the independent strategic decision of multi-game AI.According to the adjusted strategic decision,each game AI chooses the tactical decision model of independent AI to realize the united decision of multi-game AI collaboration.In united model of AI,the number of decision and decision types of multi-game AI is more,based on the consideration about accuracy of the model,this paper realized the united AI model based on BLSTM.In the process of building the model,the implicit layer of the model structure is trained and adjusted,the related parameters are fine-tuned too.The experiment shows that the team decision of the two united game AI decision-making methods can realize the coordinated operation in the combat scene,game AI's united decision-making in the battle game showed more intelligent behavior,which has practical value.