Update - Computer Generates New Doom Levels (Now With Video)

in #games6 years ago (edited)

Way back in 1993 we saw the release of Doom by id Software. Once this seminal title was available, PC gaming was never the same again. First-Person Shooters (FPS) had taken over and they were only going to improve with time and hardware improvements.
Way back in 1993 we saw the release of Doom by id Software. Once this seminal title was available, PC gaming was never the same again. First-Person Shooters (FPS) had taken over and they were only going to improve with time and hardware improvements.

What set Doom Apart

Id Software saw fit to release a level editor so that fans could create their own levels. Some of the better level designers found themselves getting job offers within the gaming community – Doom was just that popular.

Even computers can make levels

Computer scientists at Italy’s Politecnico di Milano created two artificial intelligences. One was given the task of analyzing thousands of Doom levels that fans created over the years. Once this task was completed this same computer set out to create its own levels based on what it had learned.

The second AI was given the task of also analyzing thousands of Doom levels that fans made. The difference here is that the second part involved analyzing the levels created by the first artificial intelligence.

They did what exactly?

Both AI’s analyzed the enemy placement, level size, etc to discern the best way to make new levels within the perimeters of Doom. While not perfect, there were some glitches, the levels ended up being nearly as good as what humans had been producing for decades now. Truly impressive.

The point is

If the first computer could create levels that can fool the second computer into thinking they were created by a person then the experiment would be considered a success.

Meet Generative Adversarial Network

This is the “brains” behind the artificial intelligences that were pitted against each other. Doom level building/analyzing was the method chosen to teach these two AI’s.

The end goal

The idea is to lower the cost of game development using algorithms and such to help create at least some of the levels that fans would take on.

Source: Futurism

 

What set Doom Apart

Id Software saw fit to release a level editor so that fans could create their own levels. Some of the better level designers found themselves getting job offers within the gaming community – Doom was just that popular.

Even computers can make levels

Computer scientists at Italy’s Politecnico di Milano created two artificial intelligences. One was given the task of analyzing thousands of Doom levels that fans created over the years. Once this task was completed this same computer set out to create its own levels based on what it had learned.

The second AI was given the task of also analyzing thousands of Doom levels that fans made. The difference here is that the second part involved analyzing the levels created by the first artificial intelligence.

They did what exactly?

Both AI’s analyzed the enemy placement, level size, etc to discern the best way to make new levels within the perimeters of Doom. While not perfect, there were some glitches, the levels ended up being nearly as good as what humans had been producing for decades now. Truly impressive.

The point is

If the first computer could create levels that can fool the second computer into thinking they were created by a person then the experiment would be considered a success.

Meet Generative Adversarial Network

This is the “brains” behind the artificial intelligences that were pitted against each other. Doom level building/analyzing was the method chosen to teach these two AI’s.

The end goal

The idea is to lower the cost of game development using algorithms and such to help create at least some of the levels that fans would take on.

Source: Futurism

 

Sort:  

The key though is that they need the manually created levels in order to teach the AI to create new levels. So they would need to add in generalisation in order to apply these rules to a new game.

Very true. We are far from ckmputer A.I. making a game from scratch.

Thanks for posting this, GAANs are a side interest of mine. I didn't know they were using it for level design, but it makes great sense.

It is interesting how things are progressing, that is for sure.

And not quite so far off from ai writing games as we think.

Exactly. Tech is usually a decade or so ahead of what we are seeing publicly.

Coin Marketplace

STEEM 0.16
TRX 0.15
JST 0.028
BTC 54225.25
ETH 2280.38
USDT 1.00
SBD 2.29