A star algorithm implementation problems

Question

I’m having some trouble implementing the A* algorithm in a 2D tile based game.

The problem is basically that the algorithm gets stuck when something gets in its direct way (e.g. walls) Note that it only allows Horizontal and Vertical movement.

Here's a picture as it works fine across the map without something in its direct way: (Green tile = destination, Blue = In closed list, Green = in open list)

This is what happens if I try to walk 'around' a wall:

---

EDIT

Thanks DeadMG, I rewrote the algorithm last week and got it somewhat working.

But the heuristics are still some faulty.

As you can see here it prefers a uneven path on an straight way:

The heuristics function:

float c_astar::GetH(cTILEext TILEX)
{
    short a = 0; short b = 0;
    if(TILEX.Tile.Position.x > Destination.x) a = TILEX.Tile.Position.x - Destination.x;
    else a = Destination.x - TILEX.Tile.Position.x;
    if(TILEX.Tile.Position.y > Destination.y) b = TILEX.Tile.Position.y - Destination.y;
    else b = Destination.y - TILEX.Tile.Position.y;

    float h = a+b;

    //Tie-Breaker
    float dx1,dy1,dx2,dy2;
    dx1=dy1=dx2=dy2 = 0;

    if(TILEX.Tile.Position.x > Destination.x) dx1 = TILEX.Tile.Position.x - Destination.x;
    else dx1 = Destination.x - TILEX.Tile.Position.x;

    if(TILEX.Tile.Position.y > Destination.y) dy1 = TILEX.Tile.Position.y - Destination.y;
    else dy1 = Destination.y - TILEX.Tile.Position.y;

    if(Start.x > Destination.x) dx2 = Start.x - Destination.x;
    else dx2 = Destination.x - Start.x;

    if(Start.y > Destination.y) dy2 = Start.y - Destination.y;
    else dy2 = Destination.y - Start.y;
    float cross = abs(dx1*dy2 - dx2*dy1);

    h += cross*0.001;

    return h;
}

I used the Tie-Breaker as described here. What am I doing wrong?

Regards, bryan226

Answer 1

Its always going after the cheapest F value, which seems to be wrong.

Is the entire point of F. Your heuristic is crackers. You should use something like Manhattan distance to target (ignoring obstacles) as the value of heuristic, (yours is pretty close but you need to ditch the *10) and then 1 cost/step for G. Also, your G estimate does not actually give the length of the path, because it does not include the idea that you had to take a detour, giving you quite inaccurate results.

Fixing these two problems should give you actual results for A*.