It may do what I did in that one shrine in Breath of the Wild after I got impatient trying to solve it the “right” way, and just yeet the ball into the hole by making it jump the walls. :)
I'm still a bit proud of my solution, which involved simply turning the board upside down, enjoying a completely flat surface I could easily control the ball with.
The optimum time to beat this game for me is 2 seconds.
My brother and I used to have it as kids, eventually you figure out shortcuts, and go faster and faster.
But we discovered by mistake (or rather while expressing frustration) that the quickest route ends up being a very sudden lift of the left corner, sending the ball flying above the whole board, and directly into the last hole.
With practice, you get down to one second to get to the left corner, then a second and a half to get it airborne, and for it to fly into the finish line.
I have no doubt the AI will figure it out, and probably even make it faster.
A standard demo for robot dexterity is balancing a vertical pole, which probably doesn't qualify as a 'game'.
There was also this PR stunt by Kuka, a German manufacturer of industrial robots, that challanged famous player Timo Boll in table tennis. This was almost ten years ago, not sure how it ended, to be honest.
I'm surprised no one did this maze thing earlier, seems like an obvious idea with high probability of success. And it doesn't even require much in terms of hardware.
The title of this post is not great. There's a board game with the name Labyrinth (previously named The A-maze-ing Labyrinth), which this post is not about.
I agree, I know the game you're referring to, but didn't think of it when I chose the title. I wanted to add some more context than just calling it 'Cyberrunner' without any additional information, although I'm aware that editorializing submission titles is discouraged on HN.
I'm less impressed that the AI figured it out in the end, but more by the speed with which it played in the final state.
I'm thinking one reason why human players do this more slowly and careful is that they'd get annoyed by the repetitiveness of having to start over many times and therefore adopt a more conservative strategy. Getting annoyes is of course a problem that AIs don't have.
Given that the game is always the same (layout and wind/gravity/physics don't change), it's not surprising the "robot" succeeded. This is less about reasoning and more about trial and error and memorization. I doubt you need much AI to build such a system.
15 comments
[ 4.7 ms ] story [ 45.7 ms ] threadMy brother and I used to have it as kids, eventually you figure out shortcuts, and go faster and faster.
But we discovered by mistake (or rather while expressing frustration) that the quickest route ends up being a very sudden lift of the left corner, sending the ball flying above the whole board, and directly into the last hole.
With practice, you get down to one second to get to the left corner, then a second and a half to get it airborne, and for it to fly into the finish line.
I have no doubt the AI will figure it out, and probably even make it faster.
https://youtu.be/nt00QzKuNVY
There was also this PR stunt by Kuka, a German manufacturer of industrial robots, that challanged famous player Timo Boll in table tennis. This was almost ten years ago, not sure how it ended, to be honest.
I'm surprised no one did this maze thing earlier, seems like an obvious idea with high probability of success. And it doesn't even require much in terms of hardware.
I'm thinking one reason why human players do this more slowly and careful is that they'd get annoyed by the repetitiveness of having to start over many times and therefore adopt a more conservative strategy. Getting annoyes is of course a problem that AIs don't have.