U.S. Patent No. 10,839,215: Artificial intelligence for emulating human playstyles
Issued November 17, 2020, to Electronic Arts Inc.
Filed/Priority to May 21, 2018
Overview:
U.S. Patent No. 10,839,215 (the ‘215 patent) relates to artificial intelligence (AI) which uses player data to emulate human-like gameplay. The ‘215 patent details instructions executed by one or more computing devices to generate image indicating positional information, and non-positional information, about objects in a video game. A first AI is selected from several, based at least partially on the data, is provided with the data as inputs. The AI is configured to receive the image and data as inputs, process the inputs, and generate control command outputs. The AI executes at least one command to control the computer-controlled character in the video game, during runtime of the game.
The image generated by the ‘215 patent includes a retina image which is a matrix divided into cells, each cell corresponding to a location in the video game and indicating which objects are in which cells. While processing the control operation, the computer-controlled character checks the consequences of the operation and behaves at least partially based on the consequences. The computer-controlled character plays a similar role to a player-controlled character in the video game. The ‘215 patent could make AI controlled characters more human-player-like in some games.
Abstract:
An artificially intelligent entity can emulate human behavior in video games. An AI model can be made by receiving gameplay logs of a video gameplay session, generating, based on the gameplay data, first situational data indicating first states of the video game, generating first control inputs provided by a human, the first control inputs corresponding to the first states of the video game, training a first machine learning system using the first situational data and corresponding first control inputs, and generating, using the first machine learning system, a first artificial intelligence model. The machine learning system can include a convolutional neural network. Inputs to the machine learning system can include a retina image and/or a matrix image.
Illustrative Claim:
- A non-transitory, computer-readable storage medium storing computer readable instructions that, when executed by one or more computing devices, causes the one or more computing devices to perform operations comprising: generating an image indicating positional information about objects in a virtual environment within a video game, the objects including at least one computer-controlled character; generating data including non-positional information about the objects in the virtual environment; selecting a first artificial intelligence (AI) model from among a plurality of AI models based at least in part on the image and the data; providing the image and the data as inputs to the first AI model, wherein the first AI model is configured to: receive, as model inputs, the image and the data; process the model inputs; and generate, as outputs, at least one command to control operation of the computer-controlled character; executing the at least one command to control operation of the computer-controlled character within the virtual environment; and generating instructions to output the execution of the at least one command by the computer-controlled character within the virtual environment during runtime execution of the video game.