• Create Account

We need your feedback on a survey! Each completed response supports our community and gives you a chance to win a \$25 Amazon gift card!

#ActualKrilce

Posted 30 May 2013 - 03:09 PM

Hey guys. I'm trying to create basic pacman ai(pacman character ai, not ghost ai) without ghosts for now, so he can collect all of the dots. I've managed to get it working, and now, I'm trying to print maze state for every pacman movement, so I can see how is he moving through the maze, while searching for food. And that's where I saw a problem. First few steps, and it's working nicely. The problem comes when he eats one dot, and then starts searching for a shortest way to the other one. I'm pretty confident that Manhattan distance heuristic is giving me problems. Thing is, this kind of heuristic doesn't take walls on its path into consideration, so it sometimes ends with awkward results. The problem is not G cost, because I'm never searching different terrain or obstacles, because my Pacman can only move while on a free way. Let me explain, because I'm almost sure someone's gonna point out that that's where the problem actually is. Now, in a different function, I'm making sure that pacman cannot move through walls, so I don't need to check them, and assign much bigger G cost to them. So, pacman always avoids them anyway, because he can't move through those.

I provided a picture of the thing I get between two dots.

He has already finished eating one, so I set that one as a new starting point.

P - pacman

0 - free way

1 - wall

3 - food(dot)

1. He is on a place where he has eaten the first dot.

2. Using Manhattan distance, he has calculated that that is the shortest way to next dot, and that's where he went wrong. He didn't take wall into consideration while calculating Manhattan distance, where I'll provide a code later on.

3. He saw that he can't move down because of wall, so he decided to go back to the next best node I guess, and so on.

I want to print his every step on that maze, where there can only be ONE pacman of course. Currently, it's leaving me with 5-6 P signs on that maze, which is not what I want. I think that this can all be solved out by making Manhattan distance taking walls into consideration, but the problem is none of the things I've tried were working, therefor I decided to ask for help here.

Here is the heuristic part, and I've also attached whole source code, if anyone has time to look at it. It was done in Java, Eclipse working environment.

Thanks.

public int computeGCost() {

if(this.getParent() != null)
this.gCost = Math.abs(parent.getgCost()) + 1;
else
this.gCost = 1;

return gCost;
}

public int computeHCost(Node goal) {

this.hCost = Math.abs(this.getxCoord() - goal.getxCoord()) + Math.abs(this.getyCoord() - goal.getyCoord());
return hCost;
}

public int computeFCost(Node goal) {

this.fCost = this.computeGCost() + this.computeHCost(goal);
return fCost;
}


#1Krilce

Posted 30 May 2013 - 03:08 PM

Hey guys. I'm trying to create basic pacman ai(pacman character ai, not ghost ai) without ghosts for now, so he can collect all of the dots. I've managed to get it working, and now, I'm trying to print maze state for every pacman movement, so I can see how is he moving through the maze, while searching for food. And that's where I saw a problem. First few steps, and it's working nicely. The problem comes when he eats one dot, and then starts searching for a shortest way to the other one. I'm pretty confident that Manhattan distance heuristic is giving me problems. Thing is, this kind of heuristic doesn't take walls on its path into consideration, so it sometimes ends with awkward results. The problem is not G cost, because I'm never searching different terrain or obstacles, because my Pacman can only move while on a free way. Let me explain, because I'm almost sure someone's gonna point out that that's where the problem actually is. Now, in a different function, I'm making sure that pacman cannot move through walls, so I don't need to check them, and assign much bigger G cost to them. So, pacman always avoids them anyway, because he can't move through those.

I provided a picture of the thing I get between two dots.

He has already finished eating one, so I set that one as a new starting point.

P - pacman

0 - free way

1 - wall

3 - food(dot)

1. He is on a place where he has eaten the first dot.

2. Using Manhattan distance, he has calculated that that is the shortest way to next dot, and that's where he went wrong. He didn't take wall into consideration while calculating Manhattan distance, where I'll provide a code later on.

3. He saw that he can't move down because of wall, so he decided to go back to the next best node I guess, and so on.

I want to print his every step on that maze, where there can only be ONE pacman of course. Currently, it's leaving me with 5-6 P signs on that maze, which is not what I want. I think that this can all be solved out by making Manhattan distance taking walls into consideration, but the problem is none of the things I've tried were working, therefor I decided to ask for help here.

Here is the heuristic part, and I've also attached whole source code, if anyone has time to look at it.

Thanks.

public int computeGCost() {

if(this.getParent() != null)
this.gCost = Math.abs(parent.getgCost()) + 1;
else
this.gCost = 1;

return gCost;
}

public int computeHCost(Node goal) {

this.hCost = Math.abs(this.getxCoord() - goal.getxCoord()) + Math.abs(this.getyCoord() - goal.getyCoord());
return hCost;
}

public int computeFCost(Node goal) {

this.fCost = this.computeGCost() + this.computeHCost(goal);
return fCost;
}


PARTNERS