Pong AI
Author
So, I just finished making a pong game, and I am pretty happy with how my AI turned out, but I though it might be interesting to see how other people would tackle this problem, so I could get the feel for some of the methods that are used to do this kind of thing. Anyways, here's mine, it's in java, but I think the idea is clear enough even if syntax is a bit different between different languages (the 2.1 and 2.5 are just the numbers that seems to work the best with the ball speed and screen size I was using).
computerPaddlePosition += ( ball.getY() - computerPaddlePosition ) * Math.pow( (double)ball.getX() / (WINDOW_WIDTH * 2.1), 2.5 );
Looks like your doing the standard "if ball is above paddle, move up and if it's below move down". One other way I was thinking you could do would be to figure out where the ball is going to end up right before it goes off screen and just make the CPU go there. This would make the CPU always be able to win, so you would have to throw in some probability.
I did something similar, but gave the cpu a speed restriction which got larger (ie more speed available) as the game progressed.
Degra
Degra
Well, if your paddle is moving when it hits the ball, you can slightly alter its course (putting 'english' on the ball they call it).
So your AI should consider, given the balls incoming path, the possible 3 outgoing paths resulting from -stationary bounce, english down, or english up
Then trace those 3 paths and see which one ends up Farthest from the player's paddle as he has it positioned now.
Have the AI always try to bounce the ball far away from the player's paddle.
Actually, since the ball always travels in straight lines, and paddle actions can output only 3 possible bounce trjectories. You could reduce the game to a series of simple branching decisions, then run mini-max or a bfs type algorithm, and have the AI try to force the player into a situation where he'd miss (he was on one side of the court to return the ball, but you put the next bounce on the other side so he can't possibly get there in time...)
So your AI should consider, given the balls incoming path, the possible 3 outgoing paths resulting from -stationary bounce, english down, or english up
Then trace those 3 paths and see which one ends up Farthest from the player's paddle as he has it positioned now.
Have the AI always try to bounce the ball far away from the player's paddle.
Actually, since the ball always travels in straight lines, and paddle actions can output only 3 possible bounce trjectories. You could reduce the game to a series of simple branching decisions, then run mini-max or a bfs type algorithm, and have the AI try to force the player into a situation where he'd miss (he was on one side of the court to return the ball, but you put the next bounce on the other side so he can't possibly get there in time...)
I've found a good way to make the AI beatable is to have it only move when the ball is within a certain distance of the computer's side. So even if it moves faster and reacts better the player still has the advantage of reacting sooner.
Quote:Original post by Alrecenk
I've found a good way to make the AI beatable is to have it only move when the ball is within a certain distance of the computer's side. So even if it moves faster and reacts better the player still has the advantage of reacting sooner.
That doesn't really seem fair; making the AI 'blind' to part of the board. The Human is allowed to see the whole screen, right?
How about limiting the number of motion commands the AI is allowed to input per second? Simulate it having slow fingers like a Human? But otherwise have it be as smart as possible?(see the above bounce planning concept)
P.S. did anyone remember to give the AI's paddle a max velocity just like the player's paddle?
Author
I guess part of the issue is how your pong is designed. With my pong I was using mouse input so there was no limit to how fast I could move my paddle, so I thought it was unfair to put a strict speed limit on the computer paddle, and I have seen the paddle being blind for a portion of the board and it looked a little too artificial, but at the same time I didn't want the computer paddle just flying around all the time, so that's why I made the paddle speed a function of the balls x and y position, so that it would have a smooth motion to get to the ball. I have found it looks overall surprisingly human.
I really like the mini-max idea though, It's not something I would have thought of, but it seems like a really col way to do it if you were using the keyboard for input.
I really like the mini-max idea though, It's not something I would have thought of, but it seems like a really col way to do it if you were using the keyboard for input.
Well really, you should have a max paddle speed for Both Player and AI.
even if the player whips his mouse around, have the paddle lag behind the mouse based on its max speed.
This same issue comes up in some first person shooters, where some players will cheat by turning up their mouse sensitivity, and thus be able to turn 180 instantly...
even if the player whips his mouse around, have the paddle lag behind the mouse based on its max speed.
This same issue comes up in some first person shooters, where some players will cheat by turning up their mouse sensitivity, and thus be able to turn 180 instantly...
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
