Today I worked on the prototype navmesh for the game.
Most 3d games use a navigation mesh, it's easier to store data about where you can walk in a mesh than in text arrays, the mesh can also be edited easier than rewriting an array.
Unfortunately with a random dungeon the different sections have separate navigation meshes, and in Blender you can't merge the meshes. So in this project I'm using the meshes to generate the list of walkable tiles and then they will be deleted and the data will be kept as an array. I'll be using simple neighbor checking to create a graph that can then be used with A* pathfinding.
It's all very old fashioned, but for me it's interesting to use these methods because it allows me to understand what's going on at a very basic level. I don't really expect to improve on modern AI but I can find what's good for my project and leave what I don't need.
Here's the walkmesh:
You can see the red areas are walkable, the green areas are a negative mask, I'll be using green walkmeshes to show where you can't walk. They'll be added to props to override the existing walkmesh when the prop is added.
Actually thinking about it, green for don't walk and red for walk is kind of counter intuitive. :)
Have to change that in the next update.
Here's a closer look:
The result is a a set of dictionaries, one for each tile. I'll be accessing the dictionaries separately when building the graph for A* so I don't have to access the whole set of data, hopefully speeding up performance a bit. Tiles which are out of range of the character's movement range will not be added to the graph. Characters will pathfind to a selected square inside their move radius, but it's up to the player to do long distance pathfinding each turn.
The AI will at first try to generate a path to the player using a cropped rectangle around the position of the player and the monster. If it can't reach it will expand the search area. This means that in some cases the A* routine may need to be run twice, however, in most cases it will work the first time, using much less resources than a full level graph.
I'm also thinking of doing a simple flood fill at the beginning of the AI's movement phase to find which players are accessible, or which are on islands (in a locked room for example).
Here's a quick video that shows the unfinished result:
Most 3d games use a navigation mesh, it's easier to store data about where you can walk in a mesh than in text arrays, the mesh can also be edited easier than rewriting an array.
Unfortunately with a random dungeon the different sections have separate navigation meshes, and in Blender you can't merge the meshes. So in this project I'm using the meshes to generate the list of walkable tiles and then they will be deleted and the data will be kept as an array. I'll be using simple neighbor checking to create a graph that can then be used with A* pathfinding.
It's all very old fashioned, but for me it's interesting to use these methods because it allows me to understand what's going on at a very basic level. I don't really expect to improve on modern AI but I can find what's good for my project and leave what I don't need.
Here's the walkmesh:
You can see the red areas are walkable, the green areas are a negative mask, I'll be using green walkmeshes to show where you can't walk. They'll be added to props to override the existing walkmesh when the prop is added.
Actually thinking about it, green for don't walk and red for walk is kind of counter intuitive. :)
Have to change that in the next update.
Here's a closer look:
The result is a a set of dictionaries, one for each tile. I'll be accessing the dictionaries separately when building the graph for A* so I don't have to access the whole set of data, hopefully speeding up performance a bit. Tiles which are out of range of the character's movement range will not be added to the graph. Characters will pathfind to a selected square inside their move radius, but it's up to the player to do long distance pathfinding each turn.
The AI will at first try to generate a path to the player using a cropped rectangle around the position of the player and the monster. If it can't reach it will expand the search area. This means that in some cases the A* routine may need to be run twice, however, in most cases it will work the first time, using much less resources than a full level graph.
I'm also thinking of doing a simple flood fill at the beginning of the AI's movement phase to find which players are accessible, or which are on islands (in a locked room for example).
Here's a quick video that shows the unfinished result:
Comments
Post a Comment