Most IDM: Neural Net Racecars Learning to Drive

[Joyrex]

Just needs an Autechre soundtrack

[goDel]

Also works well with

 

[diatoms]

The Two years my Family lived in Lawton, OK

1988-1989

I played Super Off Road

as much as I could

at the local 7-ELEVEN

couldn't believe a quarter didn't last longer, Ha!

Anyway, this brought back Memories:)

[Limo]

Great find!

[bitchroast]

21 hours ago, goDel said:

Also works well with

 

thought this was gonna be MUTTLEY 

Mumbly *

[Joyrex]

22 hours ago, diatoms said:

The Two years my Family lived in Lawton, OK

1988-1989

I played Super Off Road

as much as I could

at the local 7-ELEVEN

couldn't believe a quarter didn't last longer, Ha!

Anyway, this brought back Memories:)

LOL, I thought of Super Off Road Racing too and also R/C Pro Am (NES) when I saw this.

More contemporaneously, PixelJunk Racers too.

[xox]

conan osiris is still more idm

[Joyrex]

19 minutes ago, xox said:

conan osiris is still more idm

Care to share with us?

[xox]

do i really have to?!

Spoiler

 

 

[zlemflolia]

Quote

Teaching a neural network to drive a car. It's a simple network with a fixed number of hidden nodes (no NEAT), and no bias. Yet it manages to drive the cars fast and safe after just a few generations. Population is 650. The network evolves through random mutation (no cross-breeding). Fitness evaluation is currently done manually as explained in the video.

It's just a genetic algorithm that encoded this track, according to the creator, within a neural network.  Furthermore, it didn't even do it itself, but rather apparently the creator showed it how by "manually" selecting the most fit generations.  On top of that, it only learned -this- track.  As comments in the video said, the OP should have randomly generated tracks during training because this NN is completely overfit to this specific track and if you put the cars in another one they would fail completely

So while stuff like this is in general cool, this video in particular is a pretty bad example.  So this video is not very IDM, here's a more IDM one:

 

 

[Limo]

8 hours ago, Zeffolia said:

It's just a genetic algorithm that encoded this track, according to the creator, within a neural network.  Furthermore, it didn't even do it itself, but rather apparently the creator showed it how by "manually" selecting the most fit generations.  On top of that, it only learned -this- track.  As comments in the video said, the OP should have randomly generated tracks during training because this NN is completely overfit to this specific track and if you put the cars in another one they would fail completely

So while stuff like this is in general cool, this video in particular is a pretty bad example.  So this video is not very IDM, here's a more IDM one:

 

 

 

For me it was the crummy potato quality of the process in the original video that appealed to me. Yes I know that PhD's with expensive equipment can do better - but this was *fun*!

[Candiru]

Nothin but neural net

[Soloman Tump]

What was the car thing a few years back that learns to negotiate a 2D randomly generated terrain?
It grew the right size wheels and body shape to get further along the track each time.

I remember it was quite a hit on here for a while.

Edit

I did a google

 

Boxcar 2D!

 

https://rednuht.org/genetic_cars_2/

[Joyrex]

17 hours ago, xox said:

do i really have to?!

  Hide contents

 

 

Oh holy fuck I didn't realize that was the name of the Eurovision twat

[goDel]

6 hours ago, Candiru said:

Nothin but neural net

basically. he couldn't work out the genetic algorithm. a bit sad actually. not IDM worthy!