Game object control method, device, medium and equipment

Abstract

The invention discloses a control method of a game object, which comprises the following steps: obtaining a game image of the game object when the game object participates in a game, and judging whether an opponent object of the game object exists in the game image; If not, inputting the game image into the first strategy model, obtaining the action probability vector output from the first strategy model, selecting the action with the highest probability as the target action according to the action probability vector, and controlling the game object to execute the target action to realize themap exploration in the game; if so, the game image is inputted into the second strategy model, the action value vector outputted from the second strategy model is obtained, the action with the highestvalue is selected as the target action according to the action value vector, and the game object is controlled to execute the target action to fight against the opponent object. This method decomposes the control of the game object into map exploration and battle, which reduces the training time, and the model has good performance. A device, equipment, and medium are also disclosed.

Description

technical field [0001] The present application relates to the technical field of game artificial intelligence, and in particular to a game object control method, device, computer storage medium and equipment. Background technique [0002] At present, in game development and application, artificial intelligence is used in many scenarios, which is called game artificial intelligence (Artificial Intelligence, AI) or game AI. For example, in the game development process, the game AI can replace the role of the tester, and the test data can be obtained by playing the game through the game AI to realize the game performance test; , you can actively choose the game AI to automatically handle. It is also possible that when the number of players is not enough, in order to enable the game to proceed normally, some game AIs are also selected to play games with real people. [0003] At present, there are two main implementation schemes for game AI. One is the implementation scheme bas...

AI Technical Summary

Problems solved by technology

[0003] At present, there are two main implementation schemes for game AI. One is the implementation scheme based on scripts, which has the advantages of simple implementation and low computational overhead, but its disadvantage is that the behavior mode is rigid. In games such as first-person shooter (FPS), the scene is very random, and this solution cannot be applied

The other is an implementation scheme based on deep reinforcement learning. This scheme is mainly based on deep reinforcement learning algorithms and realizes game AI through online training of a single neural network model. Since this model only has a relatively large number of training samples. Good performance, but the generation efficiency of sample data in the online training process is very low, it is difficult to obtain more training samples in a short time, which leads to poor performance of game AI

Images