dumb neural network

[startoftext]

What simulation environment did you use or did you write this in open GL from scratch or something? Also just curious did you use back propagation of error or did you use a genetic algorithm to train the ANN? Very nice visual example.

[labenjam]

@startoftext

hi, I did this one with Ogre3d for graphics and back propagation for learning algorithm.

Thx.

[MrJhonnyRed]

in some corners it's trying to drift LoL it's doing a great job!

[radonas]

does it send out lines and just follows the one thats the longest?

[caiotops]

Hi. It is a very interesting demo. I would like to use it to teach my students neural networks. Is the source code available? Thanx.

[waddlerobloxxxx]

Risg fo rods?

[ailm96094]

It could be so smart you hurt it's feelings.

I think you should post an apology.

[RomSMEG]

realy dumb

[Houshalter]

How does it learn? Does it mimick examples of driving, or is it a GA.

[labenjam]

@Houshalter

That one is just learning from example.

Check the other videos where it is learning with a GA!

[MysterySultan]

Even something as simple as a car being driven amazes me, great job. :)

[signofdevil666]

all it looks like is simple AI path-finding

but good job nonetheless

[Sugardude]

No, this AI doesn't find a path, it avoids walls.

There's a neural network running the AI which takes a few samples of wall proximity as input and outputs the commands to drive the car.

[adrixshadow]

Where is the dumb?

From what I see its pretty intelligent to have completed the track

[jetsam50000]

Not really, all it is is contact sensors.

[ilovehd831]

Maybe it was sarcasm, or maybe it was in reference to the fact that the task in this situation is very simple (for neural networks).

[labenjam]

@ilovehd831

Exactly!

I meant "Simple" neural network

[salmiak911]

Are you willing to share the code?

[nickrohn93]

when u say how many neurons u arent being literal?

[gafpjiwe]

how many neurons do u have on the hidden layer ?

[labenjam]

Can't remember for this specific test but you can have an idea with this formula :

number of neurons in hidden layer =

SquareRoot ( (nb input*nb input ) + (nb outputs*nb outputs) )

So in this example I must have used at list

6 ~= SquareRoot( 5*5 + 4*4 )

[Sergiuss555]

and what the type of NN is used there?

[labenjam]

A simple multilayer feed foward, 5 inputs ( the rays), 4 outputs ( acceleration, turn L/R, brake), and 1 hidden layer.

[Sergiuss555]

hmm interesting..

[GegoXaren]

does it learn from driving several times? if so it will soon drive relay fast.

[labenjam]

no, it records how humans are playing and learns from these records.

The car itself is slow to drive, it was just a proof of concept.

[botrules]

Did the humans played on a different track?

I would be interesting NN to learn in a track and test it in another one.

[alaskajohn]

Nice job man!! Keep up the good work!

[FirebrandAndy]

hey good ben :P !!!

[dobidik]

yea, it's really dump. so slow and it stops on every bend. check out actual toy cars which're driven using neural networks. they're awesome. benjamin's network s.cks.

:) ehehe, I like to annoy benj. It's great, ben. Keep goin'.

[umutert]

wow:)