• Video 14
• Views 4 328 440
A.I. learns to drive on a pipe
A.I. learns to drive on a pipe in Trackmania, using machine-learning.

Thanks to everyone who sent me a replay on the second map !

Contact :
Discord - yOsh_85#5919

__________________________
__________________________

Some more details about the supervised learning part : I'm playing with a keyboard, and A.I. learns to predict which keys I will press for a given set of inputs. I'm recording "Left" and "Right" as 0 (no press) or 1 (press). A.I. uses a joystick, so it needs to predict a single direction value between 0.0 and 1.0. So, for each set of inputs, its neural network predicts the probabilty that I would turn left and the probability that I would turn right....
Views: 160 929

# Video

Can A.I. finish this track without crashing ?
Views 149KYear ago
A.I. learns to drive in Trackmania with NEAT algorithm, but it is not allowed to hit walls ! Contact : Discord - yOsh_85#5919 Twitter - yoshtm1 Music : thexvid.com/video/yfkI4i0Rz2g/video.html
A.I. teaches itself to drive in Trackmania
Views 3.8MYear ago
A.I. teaches itself to drive in Trackmania, using NEAT algorithm, which is a particular type of Genetic Algorithm. This algorithm is used to select a neural network with optimal weights, and also an optimal structure. Thanks Trabadia ! His TheXvid channel : thexvid.com/user/Trabadia1 More information about NEAT algorithm : neat-python.readthedocs.io/en/latest/neat_overview.html Contact : Discor...
AI plays the new Trackmania game (2020)
Views 25KYear ago
A new Trackmania game was released on July 1st. I've adapted the AI from the previous video to drive in this new edition ! Sorry about my accent, English and French subtitles are available if you want :) Some more details about the AI : I'm using artificial neural network, a machine-learning algorithm. The network has 16 inputs (15 wall distances speed) and 4 outputs (forward, backward, left, r...
AI plays Trackmania
Views 161KYear ago
Hi :) I wanted to make an Artificial Intelligence able to play trackmania for a long time. I spent a lot of time on this project and I hope you like the final result ! I'm using artificial neural network, a machine-learning algorithm. The network has 16 inputs (15 wall distances speed) and 4 outputs (forward, backward, left, right). The model in this video has 2 hidden layers (64 16 neurons). I...
3000 self-driving cars in Trackmania
Views 7KYear ago
Hey :) this is my first k-project on trackmania. In fact, it's my first big editing video on trackmania ! I tried something a bit different : generally, on this type of video, replays are driven one by one by some players. Here, replays are also driven one by one, but everything was done automatically. I wrote a script that drives automatically a replay on a map, then saves it, then starts agai...
Trackmania² C11 by yOsh (video test)
Views 4.1K3 years ago
just a test video :)
A.I. teaches itself to drive in Trackmania
Views 3.8MYear ago
A.I. teaches itself to drive in Trackmania, using NEAT algorithm, which is a particular type of Genetic Algorithm. This algorithm is used to select a neural network with optimal weights, and also an optimal structure. Thanks Trabadia ! His TheXvid channel : thexvid.com/user/Trabadia1 More information about NEAT algorithm : neat-python.readthedocs.io/en/latest/neat_overview.html Contact : Discor...
AI plays Trackmania
Views 161KYear ago
Hi :) I wanted to make an Artificial Intelligence able to play trackmania for a long time. I spent a lot of time on this project and I hope you like the final result ! I'm using artificial neural network, a machine-learning algorithm. The network has 16 inputs (15 wall distances speed) and 4 outputs (forward, backward, left, right). The model in this video has 2 hidden layers (64 16 neurons). I...
A.I. learns to drive on a pipe
Views 161K7 months ago
A.I. learns to drive on a pipe in Trackmania, using machine-learning. Thanks to everyone who sent me a replay on the second map ! Contact : Discord - yOsh_85#5919 Twitter - yoshtm1 Some more details about the supervised learning part : I'm playing with a keyboard, and A.I. learns to predict which keys I will press for a given set of inputs. I'm recording "Left" and "Right" as 0 (no ...
Can A.I. finish this track without crashing ?
Views 149KYear ago
A.I. learns to drive in Trackmania with NEAT algorithm, but it is not allowed to hit walls ! Contact : Discord - yOsh_85#5919 Twitter - yoshtm1 Music : thexvid.com/video/yfkI4i0Rz2g/video.html
AI plays the new Trackmania game (2020)
Views 25KYear ago
A new Trackmania game was released on July 1st. I've adapted the AI from the previous video to drive in this new edition ! Sorry about my accent, English and French subtitles are available if you want :) Some more details about the AI : I'm using artificial neural network, a machine-learning algorithm. The network has 16 inputs (15 wall distances speed) and 4 outputs (forward, backward, left, r...
3000 self-driving cars in Trackmania
Views 7KYear ago
Hey :) this is my first k-project on trackmania. In fact, it's my first big editing video on trackmania ! I tried something a bit different : generally, on this type of video, replays are driven one by one by some players. Here, replays are also driven one by one, but everything was done automatically. I wrote a script that drives automatically a replay on a map, then saves it, then starts agai...
Trackmania² C11 by yOsh (video test)
Views 4.1K3 years ago
just a test video :)

• Ben RUDOLPH 20 hours ago

If you can figure out the shortest distance between start point and stop, hugging the corners and driving straight through the wavy curves, the one with both shortest distance and shortest time should be the one to make it to the next round? For example, an oval track 1 km total length is less than 1 km on the inside of the curves, which is why runners have staggered start positions, even though it's not staggered at the finish line.

• Akinaro Day ago

I think You should add more variables to learning related to actual racing, like for example hitting apex on corners. So for example car that turn right, should stick more to right side of track(smaller distance to right side compared to left) and after exiting corner sticking to middle. And pushing cars that do that to next generation.

• GekoPoiss Day ago

It would be interesting to learn more about how you interface with the game programatically and run simulations in batches

• French people speaking English is the absolute worst

• Nice

• Petr Suchý Day ago

@Yosh - I think that genetic algorithm may still be used. But if you think about this - the main disadvantage of here is the decision process. Decision is made based on incomplete information. Your algorithm is solving only current turn (curve). But humans think about at least one or two turns upfront. If you are able to add at least "next" turn parameters into inputs, then it can decide better.

• Daniel Reich 2 days ago

This is super amazing and satisfying to watch! Thank you lol. Can't wait till AI tos speedruns become more possible.

• Boze De Wolf 2 days ago

This is not War Games but Race Games ... Do you want to play a game?

• Mayz 2 days ago

I didn't knew i would be this much interested about IA

• Mayz 2 days ago

Si tu utilise le "Supervised Learning" avec des runs de professionnels, ca devrait marcher mieux non ?

• Mayz Day ago

​@Yosh Ok merci Et avec l'autre méthode si tu fait beaucoup plus de génération, ils pourrait devenir ultra fort logiquement ?

• Yosh Day ago

yes c'est ce que je dis à la fin :) Mais c'est pas si simple que ça quand même, ca peut entrainer d'autres problèmes. Par exemple, en théorie, des pros vont uniquement faire passer leur voiture par des trajectoires optimales. Le truc c'est que dès que l'IA va dévier un petit peu de cet optimal, elle va commencer à rencontrer un type de données qu'elle ne connait pas, vu que les pros ne lui ont pas montré, et donc la les problèmes arrivent..

• Filip 2 days ago

A few improvments. 1. Add more inputs. Maybe dubble them. 2. More variants. 3. Let them break. 4. Let them play on moremaps to avoid overfitting. 5. Make the same variant play on like 20 diffrent maps and add the pointstogheter. This makes sure that your AI dosent reach a false best weight.

• Cet accent français

• Franseng Herre 3 days ago

Le joueur peut voir plus loin que l'IA.

• prendo 3 days ago

i love watching ai teaching himself something

• zachary hebert 4 days ago

14:23 wtf

• Adriano Kapellen 5 days ago

Maybe give some rules to the AI, that aren't allowed to be broken, like never touch the wall. That could increase the time as well

• Arne Lewinski 5 days ago

Generalization is fine. But the AI is lacking Concepts. That's why it's progressing that slow.

• Matthew 6 days ago

I've never played trackmania but I assume there's speed loss from bumping the edge, right? If so, creating a penalty for touching the wall but still favoring time overall could have been another parameter to increase efficiency.

• bob smith 7 days ago

Your approach was flawed from the beginning. Driving isn't about your distance from walls, it's about your trajectory and the upcoming twists and turns of the road you're driving. Instead of distance from walls you should have had it internally map the layout of a course and adjust it's speed and turning accordingly

• Twiggy Pan Cake 8 days ago

This gives me villain thoughts. Like cane you use it to achieve world records?

• Haulun 8 days ago

super accent francais mdr

• sare enza 8 days ago

Super interessanr, j'aurais aimé connaître les outils utilisé pour cette vidéo. Comme le language utilisé et comment tu as pû intégrer ton ai dans le jeu. Bon courage pour la suite

• Bobbies1 8 days ago

Man is teaching Asians to drive. What a hero :)))

• 希望alex 8 days ago

1. There is no evidence that such a thing ever happened or happens in nature 2. Even in your experiment it is you yourself doing the selecting, based on criteria you defined 3. No such thing as true randomness, your so called randomness is caused by factors you have determined 4. The entire progress in your experiment was overseen, guided, and designed by an intelligent mind 5. No randomness, no NATURAL selection 6. Your video is an argument for intelligent design 7. It can be argued there are more efficient ways to find the fastest path for an AI

• Rickard Rocks 8 days ago

Lovely, are you working on something new? Would love to see more.

• ScourgeHH 10 days ago

I doubt you will ever see this comment, but I was wondering, is their a way to segment generations and merge the best of each segment. For example, on some generations, we see some cars with great runs up to the first checkpoint but that wreck after that, thus, not being the best Cars of that generation, a slower car that doesn't wreck is. Is it possible to segment the data, ie, time cars from start to first checkpoint, and also from first checkpoint to second. And then pass on the traits of not the car who went the furthest, but instead the traits of the car who has the best start to first checkpoint time, combined with the car who had the best first to second checkpoint time?

• Valentin Van Rossom 10 days ago

Un peu tard comme com' mais basé l'apprentissage sur la vitesse moyenne n'est pas une meilleure solution ? Prendre celle qui à la vitesse moyenne la plus haute force à limiter les mouvements superflux ?

• SCAR aw 10 days ago

• Peristeronic 11 days ago

This is my poo…. In the pipes

• Peristeronic 11 days ago

Poop going down the pipes be like

• Hagop Ket 11 days ago

Jesus is the way and the truth and the life. No one comes to the Father except through Jesus. Jesus Christ died for you and accepting Him as your Lord and Savior and also repenting of all your unrighteous and evil acts which are called sin, you will be saved through God’s grace and have a ticket into Heaven and be saved from the lake of fire which is Hell.

• fox-fox6909 12 days ago

un petit accent français 🤔😂

• Pulmo 12 days ago

au pire parle en français et met les sous-titres english ca sera + comprehensible

• Yidir 13 days ago

un accent t'as peur

• KESHER MEDIA 13 days ago

Excellent video! 👍👍👍👍👍

• Michael Ray 14 days ago

Those cars remind me of sperm reaching the egg. Survival of the fittest.

• SSDeveloper 14 days ago

I wonder if you could use something like TMInterface to speed up time or even modify the bruteforce mode to run each run much faster

• Yan Marcos 14 days ago

Can anyone Tell me de music playing in the background?

• CD 14 days ago

If this is AI then so is this. if (1==1) {"yup";} If you do not understand why this is not AI then that is why you think this is AI. The bar with what we call things "AI" these days is so low we can say it about anything now.

13:16 he goin' backwards

• arsonistaustro 15 days ago

i wanna see an AI teach itself to finish the Rainbow Road course as quick as possible

• misterfm2189 15 days ago

you'r french no?

• Steven Claeys 15 days ago

Why only give the AI 3 options (L, R, Fwd) and not the ability to slow down as well? If the goal is to achieve better lap times with each generation, then all possible inputs need to be available for optimal results right?

• Yosh 15 days ago

You can slow down by releasing fwd, and slowing down is not usefull on this map anyway (trabadia doesn't release Fwd or use brake in his run for example)

• Targetedandfiring 16 days ago

It is possible to combine supervised learning with generational learning? Maybe the ai starts off having learned how you do it with supervised learning and after a certain point strives to do better than how it was shown?

• chari Muvilla 16 days ago

Yeah q-learning is the way to go. You get track memory + only one car per run

• Tony Animations 16 days ago

13:13 Oh my god you are simulating fluids with cars xD

• 99 % 17 days ago

Are u french ?

• Ion Borșevici 17 days ago

Your problem is in your weak vision of how to form an effective criterion for selecting the best examples for the further evolution. It is not some AI or method trouble, but it is yours one. It would be better for you to select another profession, any math-related is not yours.

• Fuzion Streaming 18 days ago

T fraçais toi , non?