Automatic Setting of Operation Assist in Video Games

Abstract

Suppose playing a video game for the first time. If the difficulty level is too high against player’s, finds it not interesting and quit immediately. It is difficult to find the appropriate assist settings for the player. In order to find the comfortable assist settings, we conducted experiments to verify the information in the screen. Data processing methods are also verified to find suitable setting for AI learning. The verification was based on the feasibility of players’ operation prediction from three points of view: “ driving line ”based on segmentation using HybridNets,“ map ”based on the coordinates on the player’s map, and “ vehicle width ” using YOLOv8 recognition results. As a result, it was found that the map could be useful to recognize the steering and the vehicle width for detecting the presence or lack of inputs. On the other hand, the recognition ratio of the driving line was lower than that of the two, and it was not useful for operation prediction. From the above results, it can be said that the “ map ” when using the coordinates of the player and the“ vehicle width ”when using YOLOv8 are useful for obtaining the data necessary for AI learning. We would like to increase the number of data and review the data processing method.