Research And Application On The Game Ai Based On Unity3D Game Engine

Benefit from the development of science and technology in recent years,the performance of computer software and hardware has been greatly improved,and related researches of Game Artificial Intelligence(AI)began to be paid attention to.Game AI,as an important part of the game,has been playing the role of improving the player experience.But developers are also aware that making a highly intelligent game AI is a challenging task.At present,the common game AI design methods in commercial games mainly include: Finite State Machine and Behavior Tree.These methods are recognized by the market because the game AI designed by them can control the behavior completely,but it is not flexible enough.As a result,a new kind of game AI emerged,which is to use Machine Learning to train a flexible game AI to meet our needs.This paper will focus on two ways of Behavior Tree and Machine Learning,and design shooting game AI with more anthropomorphic perception and flexible behaviors.Based on Unity3 D game development engine,develop a shooting game.The main research work is reflected in the following four aspects:(1)Designing a shooting game AI based on Behavior Tree method.In order to make Non-Player Characters(NPCs)embody as much intelligence as possible,this paper proposes a practical design and implementation method based on the perception ability of NPCs.Specifically applied to the game,other additional behavior nodes are designed for it to make the behavior of NPCs more completely.(2)Designing the game AI by using relevant methods in Machine Learning,which is mainly a way of Reinforcement Learning.Taking the training shooting robot as an example,the design and implementation method based on machine learning game artificial intelligence is described in detail.In addition,in order to make the training of reinforcement learning faster and more effective,methods such as curriculum le arning and curiosity are combined.The more scientific and effective training methods will be demonstrated through several training experiments of different combinations.(3)Designing and implementing a shooting game,including the game's story background,game style,basic playability.And most importantly,apply the proposed behavior tree perception system design method and the machine learning-based design method to the NPCs production process of the game.At the same time,a variety of training methods were compared and the optimal training methods were be analyzed from the data.In the design and implement of basketball player game artificial intelligence,a method of combining behavior tree and machine learning is proposed,which encapsulates the strategy model obtained by machine learning into some nodes in the behavior tree,and organically combines the two parts to complement each other.Finally,through the description of the overall running effect of the game,the intelligent behavior performance of all game AI is demonstrated.