AI design

tlaloc · Post by **tlaloc** » Wed Sep 08, 2004 11:11 am

Hello!

I want to ask you somethings. I'm trying to make an AI simulator of humans visiting some museum. I'm using Ogre for rendering and blender for modelling. The idea is to set the agents (humanoids) in the museum, lets ssay a guard and a visitor.
The visitor walks through the museum seing the things that he likes more, if the guard see something that the visitor shouldn't do he goes there and talk to the visitor.

There are a severall questions in my mind right now:

1) How can I make the agents see the 3D world and go walk from place to place?

1.1) Note that I want to settle the angle vision of the eyes of the humanoid (which could be different from agent to agent)

2) How can I make my agents and AI program completly independet of the Ogre engine?

I was thinking in putting the Ogre and Ode in the same layer if design, then there's an API which comunicates between the Ogre+Ode Layer and the AI program+agents (wich is in another layer).

Thanks a Lot! I hope that you understand my questions.

Antiarc · Post by **Antiarc** » Wed Sep 08, 2004 6:44 pm

1) Either standard pathfinding like you see in RTS games, or set "travel paths" that the the agents follow throughout the museum with some randomized variation.

1.1) Not a problem.

2) Just write your classes that do AI, and then on each Ogre frame, step your AI procedures and update the position of the models in the engine.

You shouldn't need ODE, since this isn't really player-controlled - bounding box collisions should be sufficient.

tlaloc · Post by **tlaloc** » Fri Sep 10, 2004 1:41 pm

Antiarc wrote:1) Either standard pathfinding like you see in RTS games, or set "travel paths" that the the agents follow throughout the museum with some randomized variation.

1.1) Not a problem.

2) Just write your classes that do AI, and then on each Ogre frame, step your AI procedures and update the position of the models in the engine.

You shouldn't need ODE, since this isn't really player-controlled - bounding box collisions should be sufficient.

Another question. Where should I put the pathFinding algorithm? The best choice, for me, should be in the AI classes to make it indepent from Ogre (so if I change the Render Engine It is done in the AI classes), but How can I do it indepently from Ogre? Because I need to make questions to the render engine like: what does character B sees? What objects are there? what must I avoid? etc...
I'm steel a newbie to this kind of work, so my apologies if my questions seems to stupid or to easy.
Thanks for the ODE tip

.

Post by **tuan kuranes** » Sat Sep 11, 2004 10:41 am

PathFinding algorithm could find a warm place in a PathEngine class that know more things that player should.
Some can place it in an AI class from which guard and visotor inherits...
But most pathfinding Data doesnt' belong to these classes.

Precalc' path, path already used, path that interesect each other (guard walking on visitor ?), influence map, etc....

just call a PathEngine::findPath with
start, end, vision angle, vision distance, and speed.

some links that could help
In your case : http://www.cgf-ai.com/
More general : http://www.ogre3d.org/phpBB2/viewtopic. ... 1006#41006

tlaloc · Post by **tlaloc** » Mon Sep 13, 2004 4:24 pm

tuan kuranes wrote:PathFinding algorithm could find a warm place in a PathEngine class that know more things that player should.
Some can place it in an AI class from which guard and visotor inherits...
But most pathfinding Data doesnt' belong to these classes.

Precalc' path, path already used, path that interesect each other (guard walking on visitor ?), influence map, etc....

just call a PathEngine::findPath with
start, end, vision angle, vision distance, and speed.

some links that could help
In your case : http://www.cgf-ai.com/
More general :http://www.ogre3d.org/phpBB2/viewtopic. ... 1006#41006

Thanks for the topics! It is a good help. I guess the ideal is to have some class or function like WhatISee() which returns what the agent sees. then acordingly with the agentes mood, it goes to a specific place -> PathEngine::findPath(whatIsee[0]). How can I open the eyes of my character in the Ogre frameWork?

Post by **tuan kuranes** » Mon Sep 13, 2004 4:32 pm

rayscenequeries lets you test visibility using ray test.

Pretty computational way to do so, you'll have to discretize the number of ray you want to query by some simplification.

But if you have static envirronement, you'd better precaculate what agent can see or not from one place .

The half-life paper in the topic is just about that.

tlaloc · Post by **tlaloc** » Mon Sep 13, 2004 4:50 pm

tuan kuranes wrote:rayscenequeries lets you test visibility using ray test.

Pretty computational way to do so, you'll have to discretize the number of ray you want to query by some simplification.

But if you have static envirronement, you'd better precaculate what agent can see or not from one place .

The half-life paper in the topic is just about that.

The RaySceneQueries remembers me the paper of the renderware -> blind man with a stick.
My characters needs to see the obstacles, the moving objects (to avoid collisions with another agents), and needs to now what objects are seing so it could decide to wich object is it going to interact.
Let's say a museum and the agent sees a sculpture and a painting. He goes to the painting since he prefers paintings..

Post by **tuan kuranes** » Mon Sep 13, 2004 7:19 pm

If you read the renderware paper, what they call "The 3D topology dynamic analyzer" is the mechanism that the half-life paper explains. (I think it originally comes from a gamedev or gamasutra article on visibility)

It need a static world. but people and object can move. It's occluder that can't, like walls... so if your walls aren't destructible and your museum map doesn't have moving walls, that method is the best choice (fastest above all)

tlaloc · Post by **tlaloc** » Tue Sep 21, 2004 3:02 pm

Is there any example on how to use ray queries?

Post by **tuan kuranes** » Tue Sep 21, 2004 3:34 pm

Search 3d picking in the forum.

janders · Post by **janders** » Tue Sep 21, 2004 3:37 pm

If you want an little more accurate depiction of what an agent sees, you can also use a frustum to query the scene, rather than a ray.

tlaloc · Post by **tlaloc** » Tue Sep 21, 2004 4:37 pm

janders wrote:If you want an little more accurate depiction of what an agent sees, you can also use a frustum to query the scene, rather than a ray.

How can I use the Frustum to query the scene????

janders · Post by **janders** » Tue Sep 21, 2004 8:45 pm

Get the planes from the frustum, and use a PlaneBoundedVolumeQuery.

Jon

TorqueMaster · Post by **TorqueMaster** » Mon Jan 10, 2005 10:52 am

Hi,

I am working on having my agent in the game path find their way. I have not been able to figure out how to apply the path finding algorithm A* to the my level which is a .scene file exported from 3D Max.

Any ideas,

Thanks,
Asif

Post by **tuan kuranes** » Mon Jan 10, 2005 11:41 am

It's all about pre-process you scene to something usable by a pathfinder :

Polygon Soup for the Programmer's Soul: 3D Pathfinding :
http://www.gamasutra.com/features/20020405/smith_01.htm

some AI sdk provide those pre-process step, but all that does are very costly.

WanabeCoder · Post by **WanabeCoder** » Mon Jan 10, 2005 12:28 pm

Erm.. i dont know the capabilities of ogre, but couldnt you basically define certain points (call them waypoints) in the map, like, a whole array, and use a random integer+case to decide where they would go? Wouldnt be very intelligent, but it would still be half decent.

Dont know how you would actually get them to WALK there, but hey, Im still having technical difficulties with t3h engine.

eugen · Post by **eugen** » Mon Jan 10, 2005 1:15 pm

What tuan tries to say is that u need to make your own data structure to use it as a map in which u can look whenever u need to take an AI decision! OGRE (for all i know) doesnt have anything related to path finding...anything u'll do, regardless using OGRE data structure or not, must be done by hand !!
u can actually make a wise AI just using waypoints (depending on the game u're programming) and some parameters for decision making and maybe some free action on key events (ex: spot an enemy)...but the ai implementation should come after u got your engine in place...i'm thinking right now u'll need a map editor to define all this waypoints...or maybe u'll need a script language!
good luck!

WanabeCoder · Post by **WanabeCoder** » Mon Jan 10, 2005 1:37 pm

I thought OGRE had a map editor >.>

Post by **tuan kuranes** » Mon Jan 10, 2005 1:40 pm

Yep. (my sentence wasn't very clear...)

You need to generate an special data file containing info needed by the pathfinder. (real-time analyzing a scene using raycast cannot be done real-time)

Here's a sample of how to achieve it :

Shockwave demo :
http://station-zero.com/medion/demos/pathfind.htm

Director source :
http://station-zero.com/medion/demos/pathfind.zip

The article it's based on :
http://www.gamedev.net/reference/articl ... le1841.asp

The thread I found It :
http://3d.5341.com/msg/2293.html

TorqueMaster · Post by **TorqueMaster** » Mon Jan 10, 2005 1:58 pm

Thanks all of you for the help ... Ill be implementing a dummy AI using waypoints and develop a better by creating a directed graph later ...

Asif

Ogre Forums

AI design

AI design

help regarding pathfinding