Home / Trends / War Stories: How Forza Learned To Love Neural Networks To Train AI Drivers

War Stories: How Forza Learned To Love Neural Networks To Train AI Drivers



Produced by Justin Wolfson, edited by Shandor Garrison. Click here for the transcript.

Once an upstart, that Forza The franchise is now firmly anchored in the pantheon of great racing games. The first installment was created in response to Xbox Gran Turismo, but also with a healthy dose of online multiplayer racing. Since then, it has grown with Microsoft̵

7;s Xbox consoles, with more realistic graphics and increasingly accurate physics in track focus Forza Motorsport Series as well as the development of open-world adventures (and even a journey into the Lego dimension) for the Forza Horizon Games.

If you are among the millions of people who have played a Forza racing game, you are probably well aware of the AI ​​opponents of the games called “Drivatars”. When the first Drivatars debuted Forza Motorsport In 2005 they were a major improvement over the NPCs we drove in other driving games, which often just followed the same pre-programmed route on the track. “It was a machine learning system on a hard drive that used a Bayesian Neural Network to record [racing] Lines and characteristics of how someone drove a car, “explains Dan Greenawalt, Creative Director of Forza Franchise at Turn 10 Studios, in our latest War Stories video.

In fact, the technology originated at the Microsoft Research outpost in Cambridge, England, where computer scientists began using neural networks to see if it was possible to get a computer to tell a Formula One driver by the way to identify how they were going around corners.

Drivatars have come a long way in recent years – and not always in ways that gamers will appreciate. For the first games written for the Xbox and Xbox 360 consoles, Drivatars lived entirely on the console’s local hard drive. As you lapped the different tracks in the game, the Drivatar system used this data quietly in the background to train the AI ​​you were up against. However, with the move to Xbox One, the Greenawalt team decided to use the cloud.

“Driven to Thunderhead at once, did we have a million laps in a week?” Greenawalt says. “It was crazy.”

More data, more problems

Very quickly, the team learned a lot at Turn 10 about how different players drove their cars. The tracks are not uniform – they differ in height, width, corner profile, etc. – and this complicates the way the neural network tries to generalize a racing line. It gets even more complicated when you compete against other cars. human gamers are much more aggressive with AI drivers than other humans and by Forza Motorsport 6The games began by shutting down some of the AI’s aggression to combat single-player games that turned into demolition derbies as the AI ​​tried to destroy people at every opportunity.

In this video, shot remotely due to COVID-19, Greenawalt goes into detail about how his team has overcome these challenges over the years, and even discusses the dreaded problem of the “rubber bands” where the AI- Cars go little slower so that a human player can catch up after a turn. Even if you are not a fan of racing games, you will find this game quite exciting.

Listing image by Ars Technica


Source link