Gobang Intelligent Gaming Robot Based On Self-Play

With the development of artificial intelligence,deep learning,reinforcement learning and other algorithms are widely used to solve various game problems.Artificial intelligence algorithms of various games are obtained by training neural networks.Artificial intelligence has come to a new level of development.Based on reinforcement learning and deep neural network,this paper realizes the artificial intelligence algorithm of learning Gobang game from scratch.On this basis,we build a complete Gobang intelligent game robot system combining computer vision and automated manipulator arm.We build a model based on the Actor-Critic framework of reinforcement learning,use Monte Carlo tree search to explore,use neural network as function approximator,learn value evaluation function through value-based updating method,and learn policy function through policy-based updating method.We train the neural network from scratch through Self-Play,and the whole training process does not need any human chess player experience.Subsequently,the game level of the algorithm is tested by competing with Monte Carlo tree search and human chess players respectively.We propose an attention mechanism for game games to improve the convergence speed of policy-value network.We validate the correctness and effectiveness of the attention mechanism through a comparative experiment.On this basis,a visual system is constructed by using deep convolution neural network,which realizes the function of automatically extracting chessboard features from chessboard state images and predicting the focus position in the attention mechanism.So that Gobang game algorithm can no longer rely on manual input of artificial features,nor need any other information other than the current chessboard image to make decisions.The end-to-end decision-making process is realized.Finally,we integrate the Gobang intelligent game algorithm,visual system and automated manipulator arm to realize the intelligent robot system of Gobang game.We have developed a complete process of intelligent game robot system for man-machine game,and verified the basic functions of the system through experiments,which proves the rationality and effectiveness of the robot system design.