Artificial intelligence helps to design video games

Το συγκεκριμένο σύστημα παρακολουθεί λιγότερα από δύο λεπτά gameplay ενός παιχνιδιού σε βίντεο και στη συνέχεια δημιουργεί το δικό του μοντέλο όσον αφορά στο πώς λειτουργεί το παιχνίδι, μελετώντας τα frames και προβαίνοντας σε προβλέψεις για τα μελλοντικά γεγονότα, όπως πχ τι πορεία θα διαλέξει ένας χαρακτήρας ή πώς μπορεί να αντιδράσουν οι εχθροί.

© naftemporiki.gr This system watches less than two minutes of gameplay on a video game and then creates its own model of how the game works, by studying the frames and by making …

A new tool that will accelerate the game development process will soon become of the designers and fans of video games: Georgia Institute of Technology researchers (Georgia Tech) have developed an artificial intelligence method for recognizing / learning a game engine -that is, the basic software of a game that controls everything, from character movement to rendering of graphics.

This system watches less than two minutes of gameplay on a video game and then creates its own model of how the game works by studying the frames and making predictions about future events, what course a character will choose, or how enemies can react.

To make their artificial intelligence create an accurate prediction model that would represent all events (moves, etc.) of a two-dimensional platform-type game, the team trained artificial intelligence in a single “speedrunner” video, where one player is directed directly to his goal. This made the AI ​​training problem as difficult as possible.

Researchers for the purposes of their work use Super Mario Bros and have begun to replicate experiments at Mega Man and Sonic the Hedgehog. As they have found, the game engine was able to predict video frames much more similar to those of the original game, compared to the same test when it was done on a neural network. This gave the researchers a precise general model of the game, using just the video of the game.

“Our artificial intelligence creates the predictive model without even gaining access to the game code, and makes significantly more accurate predictions of events than those of convolutional neural networks”,

says Matthew Guddal, chief investigator and Ph.D. student in computer science.

“An individual video will not create a perfect clone of the game machine, but by training only AI on some extra videos, you get something that’s close enough”

However, it is worth noting that this technique works well with games where much of the action takes place on the screen: In other games, according to Gugadil, where there is off-screen action, this system would be in difficulty.

 

Source: www.naftemporiki.gr

(Συνολικές Επισκέψεις: / Total Visits: 6)

(Σημερινές Επισκέψεις: / Today's Visits: 1)
Σας αρέσει το άρθρο; / Do you like this post?
+1
0
+1
0
+1
0
+1
0
+1
0
+1
0
+1
0

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.