Training an unbeatable AI in Trackmania

Training an unbeatable AI in Trackmania

Yosh

8 месяцев назад

12,871,404 Просмотров

Ссылки и html тэги не поддерживаются


Комментарии:

Miata Gamer
Miata Gamer - 05.11.2023 01:37

Is there some way of ,,punnishing" the AI for hitting the wall?

Ответить
Hen Bog
Hen Bog - 05.11.2023 00:39

You could program AI to use your path, which is recorded in slow motion, as a baseline to deviate from.

Ответить
Efendi Yap
Efendi Yap - 04.11.2023 19:48

When the AI went full Dr. Strange

Ответить
jstin_ 361
jstin_ 361 - 04.11.2023 18:21

Incredibly interesting and good visualization. I'm looking forward to the next video

Ответить
Raúl Jiménez
Raúl Jiménez - 04.11.2023 18:14

Amazing video Yosh. Just a beautiful example of how neural networks and reward work. Congrats

Ответить
TheXoraS
TheXoraS - 04.11.2023 15:59

Hery bro, nice project truely. But hey, petitips ici : would, could etc... le L se prononce pas du tout, jamais :)

Ответить
AKWK7985
AKWK7985 - 04.11.2023 14:36

This is insane. I would love to see the possibilities you could achieve with AI in this game.

Ответить
dikysa
dikysa - 04.11.2023 11:27

At 17 minutes you taught a bot how to drift. Then the drift reward was removed. And we further trained the bot.
Try to teach the bot with a teacher first, for example, so that it repeats after you. And then teach the bot with reinforcement.

Ответить
Dylan Edmiston
Dylan Edmiston - 04.11.2023 07:09

I am confused, did you feed the net pixel data from the game? or would that be too computationally intensive? Thats amazing though.

Ответить
Sanctus Bellator
Sanctus Bellator - 04.11.2023 06:51

The human mind took hundreds of thousands of years to evolve to what it is now, and you're giving the AI a few hours. Just think about that for a while.

Ответить
Brian Charbonneau
Brian Charbonneau - 04.11.2023 06:26

You french

Ответить
Strange Velocity
Strange Velocity - 04.11.2023 05:38

Playing against bots in the future of videogames will never be the same again. Terrifying, unbeatable and unhuman actions 😅

Ответить
Boombox
Boombox - 04.11.2023 02:32

The next step is to test the AI on the Spaghetti map

Ответить
Brad Carroll
Brad Carroll - 04.11.2023 02:24

Amazing editing. Love the visuals of the ai's all racing.

Ответить
Fritz Lauper
Fritz Lauper - 04.11.2023 01:41

Simply brilliant video! So much though and effort goining in into what seems to be a totaly meaningless endeavor. However, I am sure that the experience and knowledge gained will prove to be invaluable! The journey definitely is the reward in this case!

Ответить
MitchsMisadventures
MitchsMisadventures - 04.11.2023 01:09

Absolutely amazing story telling! I loved the use of color to easily demonstrate progression. One of my favorite aspects about this video is that you actually used your AI on multiple tracks and introduced generalization. I feel like too often people will train an AI to only be good at one specific task while ignoring other scenarios. This is one of the few videos in which you're actually training your AI and having it learn how to do something instead of just having it memorize one set of instructions. Memorizing ≠ Learning and you did an amazing job having your AI learn. Kudos!

Ответить