Sony Working to Make NPC Behavior More Believable

According to a recently published patent, Sony Interactive Entertainment is working to develop software that would use machine learning to improve NPC movements and viewpoints in video games. Multiple Sony patents published in recent months have revealed the company's efforts to create new implementations in gaming, with this latest patent adding to the growing list of Sony's development projects.

A giant in the technology and entertainment industries, Sony is a Japanese conglomerate with subsidiaries specializing in film, music, gaming, and more. Sony is well known as the parent company of the PlayStation gaming consoles, and patents continue to reveal Sony's efforts to improve video game functionality.

A newly published Sony patent describes a system in which machine learning would adapt player inputs and viewpoints to improve the movement and perspective of non-playable characters in games. The system aims to improve the current process of having NPCs solely controlled by AI utilizing a captured human play session. With the new system, NPCs would be controlled using data from both a captured session and from the live playable character. Essentially, rather than having NPC movements and perspectives strictly tied to developer input data, a machine learning system would improve the NPC by adding the player-controlled character's live movement and perspective data. Previous Sony patents relying on player input have been published, often with the goal of improving realism in video games. However, companies frequently file patents that never see implementation, so there is no guarantee this system will ever be utilized.

Patent Details for Sony Machine Learning NPC Software

Based on Figures 2 and 3 of the Sony patent, machine learning software would receive data from the playable character's viewpoint, duplicate that information, create a copied display, and use that copy to train the AI that is controlling the NPC. This would work in tandem with the existing system of using a "training set," or developer human play session, to feed the machine learning model and AI. While similar implementations already exist, they often rely solely on the playable character's perspective. Instead, this software would use machine learning to adjust inputs for the NPC perspective, thus improving realism in AI-controlled movement.

For example, during escort missions in video games, NPCs often mirror player movements, such as jumping at the exact same time. With this new software, the NPC would only jump when their perspective shows they are in the same spot that the player character was when jumping. This could help to reduce cumbersome NPC movements during escort missions, such as NPCs failing to time a jump over a gap, and could add realism to spectator views on NPC characters. Whether this machine learning model will ever be implemented remains to be seen, but it does give fans a bit of insight into Sony's R&D efforts.

As patents from gaming companies continue to spark conversation among consumers, it is key to remember that patents, trademarks, and copyright filings are never an absolute guarantee that an idea will be brought to fruition. That said, many players choose to utilize these publications to get a general idea of what Sony and other companies may be working on.

Related Forum: PlayStation Forum



"Sony Working to Make NPC Behavior More Believable" :: Login/Create an Account :: 2 comments

If you would like to post a comment please signin to your account or register for an account.


great to see them getting stuck in to improving npc behavior


I truly think gta 6 npc will be insane and much more realistic