AI Learns to Drive an F1 Car

My first video on CZcams taught an AI to destroy the game Stack's world record. After about 3 years, I wanted to create an AI which would successfully learn to drive. I would accomplish this using Neural Networks and a genetic algorithm.
Machine learning is a topic I've always been very interested in. Seeing how it can be applied to games is especially interesting. These networks can be used to teach game AI for varying difficulties.
To make it interesting I decided to use Formula 1 cars.
Make sure you subscribe!
Twitter: / skrelo
McLaren Formula 1 3D Model (Excalibur):
Link: skfb.ly/onIvF
● Watch my other videos
📹 Remaking Trouble In Terrorist Town | Indie Game Devlog #0 (Skrelo)
Link : • Remaking Trouble In Te...
--------------------------------------------------------------
📹 Making my First Mobile Game in ONE Day (Skrelo)
Link : • Making my First Mobile...
--------------------------------------------------------------
📹 5 Years Of Learning Game Development (Skrelo)
Link : • 5 Years Of Learning Ga...
--------------------------------------------------------------
📹 How I Made Minecraft In 1 Week Using Unreal Engine 5 (Skrelo)
Link : • How I Made Minecraft I...
--------------------------------------------------------------
● Music
♪ Starter [8 bit Dub]
Link : opengameart.org/content/start...
--------------------------------------------------------------
♪ Chill Lofi Inspired (omfgdude)
Link : opengameart.org/content/chill...
--------------------------------------------------------------
♪ Ooame (TAD: / tadon )
Link : opengameart.org/content/ooame
--------------------------------------------------------------
#gamedev #unrealengine #machinelearning #geneticalgorithm #neuralnetwork #ue4 #ue5

[@Unity_and_Peace]

can you do a full tutorial on how to make this pls pls pls

[@raccoon_05]

Any chance for a tutorial on this? This is amazing!!! 🤩👍

[@miencyfratgyueev3]

i dont really see why your videos arent blowing up more. you have great audio and narration as well as interesting video ideas and good explanations. youtube algorithm slacking ong

[@Skrelo]

Hopefully soon! I'm enjoying the grind though

[@masteronepiece6559]

Because he is not diving deep in details.
I mean what type NN are you using and why.

[@MrPaPaYa86]

since the premade track has curbs on boths sides in all its lenght, you might avoid the boring process of setting up walls by flagging cars running on curbs or surpassing them. I don't know UE but it's probably doable giving cars a collider underneath and checking for collisions with curbs. The collider can be smaller than tires width so they don't flag until the curb is fully overstepped

[@42meow]

"...computer sounded like it was going to war..." that's beautiful

[@midwestentertainment976]

This is wild and easily one of the most interesting explanations of so much I’ve been curious about.

[@bobross38]

Awesome 👏 now I can finally watch my Honda Accord 2004 beat a Ferrari in a race, jokes aside, this must have been a lot of work including making and editing this video! Thank you ✨😃

[@mabre4705]

Really entertaining video to watch.

[@omkar6582]

Great Job Mate. I hope this video blews up too. I've started learning Python as my starting language thanks to your support. Also I'm excited for the game's development. Oh yeah it's 00:24 AM so I'm gonna sleep now. Saw your video notification so I came running (lies I took few minutes T-T). Anyways good luck for future project and videos.

[@Skrelo]

Appreciate that! Get some sleep!

[@tirterra1222]

nice video as always

[@remilacroix122]

I would love to see the ai getting seriously quick

[@JamesPound]

What ML tools did you use for this? Is it something anyone can use? Unity has ML-Agents for example.

[@-Pls-]

Ai go brrr- meant vroom

[@Bub4597]

I'm doing something very similar, how did you make the line tracers stop when they hit the wall? For me they continue through as the color green and fill up the screen

[@hanshauser1485]

Hi Skrelo! I was just wondering, what does your fitness score look like? I programmed my own NN with Genetic Algorithm, and the AI does learn, however, it doesn't seem to try and slow down, or react to the walls near the car. My fitness score at the moment is just distance travelled along the track. Thank you in advance!

[@yannmassard3970]

Using tensor flow ? How did I go from c++ to python ?

[@allanbru6068]

Maybe you could use random spawn points along the track to improve performance?

[@paulsoaresjr12345678]

PLEASE TELL ME THERE IS A TUTORIAL WERE IT IS SHOWED HOW TO MAKE THIS?

[@liamlife9940]

Its an amazing video, can you do a video tutorial code to build this ai?

[@Leo-if1ty]

hi, i am implementing such a thing. I got stuck in passing line data to the genetic algorithm. Could you give me some advice?

[@Skrelo]

In Unreal Engine? Make sure your neural network function accepts an array of data, my raycasts returned a float value for the distance to the wall, I'd put all of the distances into an array, and would pass that in as a parameter to the neural network function!

[@sral2649]

What language did you use?

[@Skrelo]

C++ and Blueprints in Unreal Engine!

[@NR_5tudio]

BRUH pls make a tutorial about neural network

[@bpwnz360]

feels like a half finished project

[@Rabbit14]

it didn't even learn or follow a proper racing line...............