Method and device for controlling an agent

Abstract

The invention provides a method or device for controlling an agent. The method includes: obtaining state information of a virtual environment; based on the state information, selecting an action type to be performed by an agent in the virtual environment from predetermined action types through a reinforcement learning neural network; extracting an action from the state information attribute; and controlling the agent to perform an action in the virtual environment based on the selected action type and the action attribute. The beneficial effect of the invention is that the intelligent body can learn more efficiently through the deep reinforcement learning network, and finally obtain more powerful performance.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a method, equipment and storage medium for controlling an intelligent body. Background technique [0002] As a major industry in the 21st century, cultural industries using images such as movies, comics, and games are booming. Among them, the game industry, as a high-value-added cutting-edge knowledge-based complex industry, needs innovation and technological breakthroughs to achieve healthy and orderly development. [0003] First-person shooter (FPS, First-person Shooter) is a type of shooting game based on the player's subjective perspective, that is, players no longer manipulate the virtual characters on the screen to play the game like other games, but experience it personally. The visual impact brought by the experience game of the environment enhances the initiative and realism of the game. Early first-person shooter games generally gave players the stimulation of s...

AI Technical Summary

Problems solved by technology

However, this end-to-end reinforcement learning algorithm cannot handle huge decision spaces, nor can it extract high-level semantic objects in games, such as enemy positions and spatial information

Images