As you well know, nowadays the neural networks they are used to accomplish everything from understanding models to predicting how certain decisions are made. But it would seem that the latter can also perform other important tasks within the videogame sector, such as for example create whole game worlds. According to what emerged from a recent patent by EA, soon neural networks could greatly simplify one of the most time-consuming and time-consuming processes for development teams.
As can be seen from the image, which you can easily find at the bottom of the article, the soil generation system patented by Electronic Arts starts with the insertion of a graphic sketch containing the basic data of the surface to be created, such as theelevation. From there, the neural network assigns the most appropriate terrain type to each point, creating mountainous regions for the higher elevations, then moving on to plains or hills for the lower ones. According to the data obtained, the results would not be very far from some of the elaborations created by the users for how Minecraft o Animal Crossing: New Horizons.
But that's not all, since the network would also be able to add specific topographical features such as caves, ponds and distinguish between different types of terrain. For example, an area marked as “flat” in the dataset could easily become a desert, a meadow or a forest. It all depends on the decision computation of the neural network, which once having analyzed everything, determines the best result. Furthermore, the latter is able to create a great variety of biomes, identifying the boundaries between each of them, creating geographical features based on real-life data.
According to what emerged, the patent was filed for the first time by a team of EA in 2019, even if it was officially granted only on February 16, 2021. Currently the US company has been forced to temporarily close, due to the strong winter storms in Texas, so at the moment it is not clear which games will be exploited this wise use of neural networks.
