Simple AI Tutorial with NEAT-python

Let's make a simple AI with NEAT-python.

[gumbo]

its always the small channels that explain it the best. Hard to find sometimes though

[livcool]

Thank you for the video, I'm going to try this out now

[syedsiti]

Thanks, I download your file and simulate. Awesome.

[turdle]

Y do u calculate the back left and back right position of the vehicle for collision? It's a rectangle :D

Also, couldn't you use the radar as your detection as well. Saves you a method call. As long as the line isn't smaller than 1, u haven't collided

[ai.simplified..]

The bad part is this machine can survive just in this road or envirment, not in all possible envirments

[pratyushpatnaik]

Can I get your contact info, I had some doubts, like discord..

[aryaanand]

hey cheesy AI, I made a flappy bird AI but the birds don't learn anything over the generations I also tried your genomes[i][1] solution but that does not work either. If you have time please look at the code and help me figure out what is wrong. Thanks in advance.


also made a geometry dash version to see if something was missing in the flappy bird game, but same results

[lostecho]

I searched for this only because of curiousity even though im not interested in AI but I just want to point out that your explantation is very awesome and helpful hope you continue this amazing work you did! ❤

[KorekPolski]

Thanks, I'm doing a diploma about Hyper-NEAT and this helped me to understand the usage of NEAT. Is nice.

[pseabrook]

Great video! Going to dive into your code for sure.

[kid-miku-kun]

When I try to play the code it will only show one car? I can see in the config.txt that there should be 30 cars.... but there's only 1? Also... it says nothing about generation and only says to press 'm'?
Help?

[Sstevewong]

good .... its need to generate generation :16, as I change some road map

[igorbronzato]

I changed the code a bit and i got 8~7 cars destroying game

[AshburnArmorerDan]

Firstly, nice video, and thank you. You can drop all but the 45 degree inputs and go with one output (negative for left, positive for right, 0 for no steering change). It solves it just as well and is a little more logical for the tanh function.

You could add lap counters, a gas and brake system, and then change the motivation to time instead of distance. As it is, it rewards wobbling as much as moving straight even though the straight line is faster.

In any case, nice work on producing a functional example. Shout-out to past you from 3 years ago. I would have loved to found this earlier.

[suryanshankur]

Dude what ide is that ? It looks very nice

[OleguitoSwagbucks]

try to play a race game with no brakes/downgear

[CHARDONNAAAAY]

Hello, I am a university student and was wondering if I could use some of this video for an assignment on neural networking?

[billygold]

Great Video... When i wanted to add more inputs, I changed the config file to be 7 inputs but when I do this, it causes a runtime error. How can I add more inputs?

[willemmichielssen]

Hey, i have a similar program with cars and those sensors but when i run my program, the ai does not copy the best performing car like in your program, sometimes a car gets further but its always one who follows the last best and it is very rare so i think its just the ai searching a bit in the direction of the best car. Most of the cars always crash in the beginning, even if there is a car who gets very far. Do you have a solution to make the other cars be like the car with highest fitness but with a small change?

[dubiousinsights]

Cheesy, I'm planning to use the code in your Github as a starting point for a self driving Video I'm working on. I will of course reference your work. Let me know if you have any objections !