Research On The Hitting Strategy Of Billiard Game Based On The Search Of Maximum Expectation

With the research of machine game,many search algorithms have been explored,especially in discrete environment type chess match,and many mature search algorithms have been applied to various chess AI algorithms.However,in the field of continuous state space,the search algorithm is not easy to be applied.There are only some theoretical researches,but no practical tests.This dissertation studies how to apply the search algorithm to the billiard hitting strategy under the continuous state space in the context of billiard matches.In table tennis competition environment,the state and action Spaces are continuous,want to use the classical search algorithm migrated to this scenario is difficult to achieve,unlike chess and billiards game,the player in playing chess,chess action always can get the situation of a certain change after the results,and table tennis ball movement is made up of several continuous variables,so do to hit the ball the ball after the actual trajectory with the expected trajectory is a bit error,it also research on the strategy of table tennis ball has brought certain challenging.According to the characteristics of billiard game,this dissertation puts forward the step generation function,evaluation function and search algorithm suitable for the billiard game environment,so as to study how to apply the search algorithm in the billiard ball hitting strategy.A billiard simulation system for search algorithm training is implemented to provide experimental training environment for search algorithm.The research content includes the following aspects:(1)A billiard simulation system suitable for search algorithm training is implemented.Since the search algorithm requires constant recursive calculation,the simulation system needs to be able to simulate any given game situation and selected action to get the game situation after hitting the ball.Specific work include,converting the reality table tennis movement in the scene to dynamics model and the mathematical model,design the simulation module complete game of table tennis,to simulate the process and rules of the game table tennis game engine to implement in the Java programming language,and implements the visual component usability validation was carried out on the table tennis game engine.After the verification,the billiard game engine is expanded into a Web application and deployed to the server,which can not only satisfy the remote interface call of algorithm program,but also facilitate the call of programs written in other languages.At the same time,the simulated data is recorded in the database for data sampling and analysis.(2)Propose a hitting strategy suitable for billiard game.Firstly,aiming at the difficulty of traversing all the optional actions in the continuous movement space of billiard,a footwork generation function which can search for a finite set of optimal hitting actions from the current situation is proposed by using the domain knowledge of billiard and the knowledge of probability theory.A function to calculate the probability of hitting success for each action and an evaluation function to score the current situation are put forward.On the basis of the above work,combined with the idea of reinforcement learning and search,this dissertation proposes a search algorithm suitable for table tennis match,which aims to find out the hitting action that can help players win the match as much as possible according to the current situation of table tennis match.(3)An experiment was designed to simulate two players using greedy strategy and the search algorithm proposed in this dissertation to conduct a virtual competition.