MariFlow – Self-Driving Mario Kart w/Recurrent Neural Network

46 thoughts on “MariFlow – Self-Driving Mario Kart w/Recurrent Neural Network”

1. Did you really meet Elon Musk? That photo looks great! 🙂

2. I'm no parent but theres a sort of parent child relationship between mariflow and him

3. We need a speedrunner to input data into this neural network. That would be interesting.

4. Train it with a TAS 😉

5. Imagine a speedrunner training mariflow

6. We seriously need a neural network mario kart tournament.

7. Could you explain what is the input to your neural network? Are you using the same pixels that are displayed on the screen with the full resolution?

8. fuck mario i am doing this for fortnite… xD

9. I tought, no, it cant be number 3, number 3 is too bad for a human :'D

10. How do you turn it into the ultimate Mario Kart player?

11. Great contribution !!

12. Thank god I know this now in 2018, my future is bright!!!!!)

13. You sound like C.P Grey.

14. I just love SethBling's neutral voice xD

15. @ 3:42
    "I recorded 15 hours of training data."
    Otherwise known as playing the game. xD

16. can anyone explain how do this please I need learn

17. I created a rocket league bot that was heavily inspired by this video.

18. Did you also try to make your neural network learned by itself like in like MarI/O project ? I kind of think that it would only cost you minor changes (gives a rate to a game session based on the time to complete the tour and the ranking). Or are there larger difficulties in this case ?

19. So cool man! Great work!

20. I have no idea what's going on, but I'm thoroughly entertained

21. your dad wasn't hopping

22. wish i was as math oriented like this guy

23. How much money does a guy like this make?

24. I have no idea how this works but this is cool.

25. hi nice work ??

26. I guessed which one was driven by the dad!

27. Great stuff!

28. you rock ,my fan

29. 15 hours…

30. Just want to ask what source(book, paper) you were using to learn recurrent neural networks?

31. I bet that this will really useful for self-driving cars, sadly we don't have banana peels to try easy the road. 😀

32. Did you ever run an AI to see how good it could be overall at this game? I would love to see some epic time trial runs or 150cc races where this thing just destroys whatever a human could do. I can't count the number of hours I spent playing this game as a kid racing around with Bowser doing time trial runs to drop my record time .05 seconds lower than a previous run. I'm pretty sure that whatever my time was this ai could shave off 2-3 full seconds.

33. Very cool, makes me think of markov chains and cybernetic feedback.

34. Train it to mine bitcoin

35. Woah, free code! thanks man, this stuffs super interesting

36. I understood some of those words

37. This is Q-Learning i think.

38. Saying sigmoid and showing tanh you sneaky bastard.

39. for the 2nd time how can I learn about this field?..I can learn from scratch if I have to

40. Sooo mariflow learns from player's play style while marI/O learns by generation evolution, what if we combine both? Learn from the player's play style/ footages then evolve from that?

41. Some complex shit right here

42. Skynetio ?

43. I wonder if this could be applied to procedural generation to make a 100% ai designed game that is still well designed.
    Like, a human level designer and the ai designer could work in this back and forth and help each other to train the ai to make levels itself.

44. Around 4:50 , how do you make such interactive sessions? Do you have a switch button that switches from the model predicting buttons to start recording training data??

45. Does it work in win10?

46. could you teach a neural network to beat kaizo?

Comments are closed.