Training method and device for intelligent game agent

Abstract

The invention discloses a training method and device for an intelligent game agent. The training method comprises the steps that an own state, an observation state and an output behavior of a player character are encoded to obtain player data, and an intelligent player agent is constructed; a pre-established initial NPC is adopted to perform interaction training on the intelligent player agent, and parameters corresponding to a player behavior strategy are updated according to parameters of the intelligent player agent after training; according to the parameters corresponding to the player behavior strategy, the trained intelligent player agent is adopted to perform interactive training on an intelligent NPC agent, and according to parameters of the trained intelligent NPC agent, parameters corresponding to a NPC behavior strategy are updated; and the parameters corresponding to the NPC behavior strategy is iteratively used for optimizing the parameters corresponding to the player behavior strategy, and the parameters corresponding to the NPC behavior strategy are optimized by using the parameters corresponding to the player behavior strategy until the NPC behavior strategy is stable. The problem that in the prior art, it is difficult to obtain the interactive data with a player in the non-player character training can be effectively solved, and the training effect of the intelligent non-player agent can be effectively improved.

Description

technical field

[0001] The present invention relates to the technical field of artificial intelligence, in particular to a game agent training method and device. Background technique

[0002] With the prosperity and development of the modern economy, the continuous progress and improvement of electronic games has now become one of the main ways of leisure and entertainment for people. Behavioral decision-making of non-player characters in games is an important factor affecting game quality and user experience. The realization of traditional game artificial intelligence is realized through a state machine or a behavior tree, through which the agent adopts different strategies in different environments and states. However, as the game state increases, the environmental factors become more and more complex, and the traditional method is complicated and inefficient.

[0003] At present, reinforcement learning has been widely used in many fields such as intelligent control, int...

Images