Thursday 4 August 2011

A* Pathfinding Tutorial : Part 2

Welcome to part two in my series on pathfinding. In this tutorial we will be looking at the actual A* algorithm before we actually start implementing it in code in the next tutorial.

This tutorial will be mostly theory based with very little coding so please stick with it as it is really important to understand the actual algorithm before trying to translate it into code!

Before we can start looking at how the A* algorithm actually works, we will need to define a few key terms that will make describing the algorithm a lot easer :

  • The G Value – Each node will have an associated G value, this is how far we would have to travel to get from the starting node to this node.
  • The H Function (The Heuristic Function) – This is a method that will help us to estimate how far we would have to move to get from one node to another node “as the crow flies”. The method will not take into account any obstacles which is why it will only be an estimate.
  • The F Value – Each node will also have an associated F value. This is a rough estimate of how far we would have to travel to get from the starting node to the end node if the path crosses this node. To work out an F value we can do the following : F = G + H – it is basically how far we will have already had to travel to get to this node add a rough estimate of how much further we need to travel.
  • The Open List – This is basically a big list of all the nodes that are being considered to be included in the final path.
  • The Closed List – This is a list of all of the nodes that have already have been looked at, we will use this to make sure we don’t consider each node more than once.
  • The Parent Reference – Each node will have a reference to the node that added this node to the Open List. When the algorithm has finished, we will use the reference to the node’s Parent to trace the final path from the end node to the start node ( I will cover this in more detail later ).

So now that we have all of the definitions out of the way we can look at the actual algorithm, if at first it seems to make no sense, don’t worry about it, just try and keep reading it through and then when we come to actually implement it in code it should make a lot more sense. It will normally take a couple of implementations to fully understand the algorithm.

So basically the algorithm works like this :

1) Clear the Open and Closed Lists and reset each node’s F and G values in case they are still set from the last time we tried to find a path.

2) Set the start node’s G value to 0 and its F value to the estimated distance between the start node and goal node (this is where our H function comes in) and add it to the Open List.

3) While there are still nodes to look at in the Open list :

a) Loop through the Open List and find the node that has the smallest F value (We will refer to this node as the Active Node).

b) If the Open List empty or no node can be found, no path can be found so the algorithm terminates.

c) If the Active Node is the goal node, we will find and return the final path.

d) Else, for each of the Active Node’s neighbours :

i) Make sure that the neighbouring node can be walked across.

ii) Calculate a new G value for the neighbouring node ( This will be the Active Node’s G value + the cost of moving to the neighbour node – I will discuss this later)

iii) If the neighbouring node is not in either the Open List or the Closed List :

(1) Set the neighbouring node’s G value to the G value we just calculated.

(2) Set the neighbouring node’s F value to the new G value + the estimated distance between the neighbouring node and goal node.

(3) Set the neighbouring node’s Parent property to point at the Active Node.

(4) Add the neighbouring node to the Open List.

iv) Else if the neighbouring node is in either the Open List or the Closed List :

(1) If our new G value is less than the neighbouring node’s G value, we basically do exactly the same steps as if the nodes are not in the Open and Closed Lists except we do not need to add this node the Open List again.

e) Remove the Active Node from the Open List and add it to the Closed List

And that’s all that this algorithm does, it basically just takes the node that is ‘closest’ to the goal node and keeps hoping across the neighbouring nodes until it reaches the goal node.

In the next tutorial, we will be going through this algorithm step by step and implementing it in code, but in the mean time I would highly recommend that you try and code this by yourself first! I know it may look pretty complicated but if you break follow it through step by step it is actually pretty simple.

The first thing you will need to do is add the new fields to the SearchNode class, a couple of floats for the F and G values and a SearchNode property for the Parent field. Then our Pathfinder class is going to need two new lists to represent the Open and Closed Lists.

The only thing that won’t be easy to code is the code that will return the final path, but you can always just miss that step… ;)

I hope you have found this tutorial useful and please let me know if anything in it was poorly explained or not explained in enough detailed or even is something was missed out completely and I will do my best to fix it.

Thank you for reading!

I would just like to mention a few of the sources that helped me when I was learning about A* :

16 comments:

  1. Also want to point out that if you ever have any questions concerning any AI-related topics including pathing, I have always received kind and helpful advice from the users at aigamedev.com

    ReplyDelete
  2. This comment has been removed by the author.

    ReplyDelete
  3. Hi, thxs alot for all of those tutorial that's help me alot.
    But i follow ur tuto since the start and i still got a little problem. When i kill a monster or when a monster reach the end i don't win any gold and i don't lose live. I look ur code and mine but i don't see where is my mistake did u have any idea ? (I can send u my project if u want :x).

    Thxs in advence

    P.S. Sorry for my bad english :s

    Edit : Nevermind i fixe my problem. I just add Player player to my update WaveManager méthod (public void Update(GameTime gameTime, Player player)) and same thing in wave & game1 and that work now

    ReplyDelete
  4. Nice, I hope you will post the 3rd version soon! This (http://www.phstudios.com/forums/index.php?/topic/1909-towerdefense-astar-algorithm/) tutorial helped me a lot, it's a great site with lot's of xna tutorials. I hope someone has any use of it!

    (This guy is also making a towerdefence tutorial =P)

    ReplyDelete
  5. Something I would like to suggest to you as either one of your tutorials or just something to keep in mind is the use of priority queues instead of lists. In pathfinding on A* and your open list gets huge, the overhead for finding the smallest F-value can get quite large. Using a priority queue: http://en.wikipedia.org/wiki/Priority_queue can greatly reduce the this overhead and provide exponential performance boosts.

    ReplyDelete
  6. I can't wait until the third installment. Any estimates on when?

    ReplyDelete
  7. When will the next part be released??

    ReplyDelete
  8. Thanks for these tutorials, I was pulling my hair out on how to implement pathfinding for a small maze game. I created a level similar to the way you did it, only I had two separate arrays for the horizontal and vertical walls since they would be implemented differently, and created a levelpath based on those 2 arrays. I understood the theory of it before, but didn't really get how to implement it properly with C#. Your first tutorial was just a godsend, i couldn't believe I missed such an obvious way :D. Thank you.

    ReplyDelete
  9. This comment has been removed by the author.

    ReplyDelete
  10. One question about the steps though, in step 3 C and D, you say if the active node is equal to the target node, return, if not, continue checking the neighbours. Does that mean that in code it basically just loops through the neighbours part over and over again ? ( which is what I have at the moment ) Because that way, it's never setting a new "activeNode" its just hopping through the actual neighbour nodes, continuously.

    ReplyDelete
  11. Are you still working on this?

    ReplyDelete
  12. Same question here,we are waiting for the part 3. Will you keep going?

    ReplyDelete
  13. Hi, great tutorial....can you do a tutorial in gamestate after part 3?

    your the best keep it up :)

    ReplyDelete
  14. so heres my attempt at implementing A* completely
    (also, I found the traceback to start part pretty easy, unless I did it wrong lol)

    Level.cs (map.cs): http://pastebin.com/R6Pz5evn
    Node.cs: http://pastebin.com/ubT6iF2G
    Pathfinder.cs: http://pastebin.com/hzdLN3Ge

    I JUST got done writing it. Now im going to go test it lol

    ReplyDelete
  15. Never mind. No matter what I do I can not seem to get my code to work. Most of the the time I end up in an infinite loop until the closed list runs out of memory...

    Here is my most recent attempt at the pathfinder.cs class

    http://pastebin.com/7UbbUBcY

    ReplyDelete
  16. I will have a look tonight and see if I can find where you went wrong!

    ReplyDelete