The Design And Implementation Of Game AI Based On Profit Of Behavior Tree

At present,most game companies,especially card game companies,usually use the infinite state machine and behavior tree to build the development system of game AI(Artificial Intelligence).The above system is easy to build in the early stage of a game,but with the frequent upgrades and modifications of gameplay,the AI system has been unable to meet the game's rapid iteration upgrade.Each modification will inevitably cause a large number of work to change behavior or state.Moreover,it can't be update in time after the bug is fixed.The expansion is very difficult when the number of states in a state machine is large.And the maintenance cost in the later stage is also very high.These have seriously reduced the development efficiency of game AI.In particular,the AI system which is built by state machine and behavior tree has rigid behavior selection and single logic,which reduces the playability of the game.In addition,it can't choose better strategies in the choice of behavior strategies,which limits the application of game AI.This paper aims to study a new AI development methods for game.Its core is to quantify game behavior with behavior value.Define a set of basic behaviors and give them appropriate value.Generally,the behaviors in the game can be combined by this group of basic behaviors,and the final value of the quantitative behavior can be calculated by collecting the benefit which generated by the basic behaviors.So the behavior that is most valuable can be visually selected in the decision-making of many behaviors.The basic behaviors are independent of each other,and the coupling between the game behaviors is low.The top-down decomposition of each game behavior can be seen as a behavior benefit tree which can not only cope with the complex and changeable state by calculating the value of quantitative behavior,but also has no behavior tree's tediousness.So it is easy to iterate and can improve the development efficiency.Based on the deep understanding of game AI technologies such as behavior tree,this paper creatively uses the way of quantifying behavior benefits on the basis of its tree structure to replace the choice of behavior logic,and designs an AI model based on profit of behavior tree,and experiments show that the model has obvious superiority in development efficiency and behavior strategy selection.Secondly,through the combination of the ELO point matching algorithm and the behavioral benefit tree model,the AI can adaptively and dynamically adjust its behavior strategy to match the player's level.Simulation experiments have proved that the playability of the game has been significantly improved.Final,in the system of game AI framework which is based on behavior benefit tree,the data is innovatively isolated in the host program,and the logical calculation is in charge of the Lua virtual machine.This design is to realize the separation of data and logic.By this way,it is very convenient for real-time adjustment of AI behavior logic and online hot update.