AI Learns to Fly in Trackmania



I trained an AI using reinforcement learning to beat one of the craziest world records in Trackmania (A12)

Support the channel on Patreon:

Contact:
– Email: yoshtm.yt@gmail.com
– Discord: yosh_tm

Sound design and mix by Samuel Badel:

Music:
– Also Sprach Zarathustra – Richard Strauss
– Sonata No. 14 In C Sharp Minor, Op. 27 No. 2 ‘Moonlight’_ I. Adagio Sostenuto – Ludwig van Beethoven
– One to Save Us All – Bonnie Grace
– Ventus – Christoffer Moe Ditlevsen
– Rise of the Velcro – Gabriel Lewis
– Horisont – Owen Meyers
– Sumerian Paradise – Dew Of Light
– Orthosie – Ben Elson
– Swan Lake Dance of the Little Swans Act II – Pyotr Ilyich Tchaikovsky
– Dianna Beach – Dew Of Light
– Hiding Your Reality – Kevin MacLeod
– Gargoyle – Eden Avery
– The Flux Beneath It All – ELFL
– Waters Will Flow Again – Gabriel Lewis

source

48 Comments

  1. The idea that it plays the game without being able to see it is strange. Just using parameters and making educated guesses about what yields the highest reward. I picture a blind man playing the game while strapped to a chair; if he does well, he receives food; if he does poorly, he receives a slap across the face.

  2. You know the solution and are just guiding something to find it. Would be notable if an ai "found" something new and didn't just brute force it.

  3. Is an approach where you let the bot learn the game physics possible? Wouldn’t that turn the runs into a math problem instead of try and error?
    If the bot knows how the car accelerates on each surface and knows the position of checkpoints would it be able to solve the fastest way on itself? Just curious I don’t know how this stuff works.

Leave a Reply

Your email address will not be published. Required fields are marked *

You might like

© 2026 Cantinho do Vídeo - WordPress Video Theme by WPEnjoy