Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[3D Rad Developer - Fernando]

I think this specific AI problem alone would deserve its own big topic 

[Mike_Hense]

i agree... 

whole heartedly...

--Mike

[n_iron]

Hi yes I agree an aircraft AI would require many different controls.

I thinking this AI will be a linkable object to the Character Object controlling it as you can in script.

a scene might consist of rigid objects, enemy or player characters, car object, cover positions and target positions. these would be linked to the AI.

now the A* Script I worked with required a walkable map and the script returned the path.

recently I tried to define map using the scan object but didn't get far due to lack of time. could the scan object be used as a separate or a part of the AI object to provide a walkable map.

my theory was to make a array of the visible area in front of a character all walkable, then fill it with non walkable objects and targets using the scan object. this data can then be used to pursue the target or flee depending on the mode. if the target is not visible then the character would go to search mode choosing a random position in the walkable map determined by a search radius, once that position has been reached they would then search again or return to its post. the characters path could also be added to the walkable array data so the enemy can pursue even if out of site. optionally this path could decay if it get to old. when a AI character is in a cover position you don't want them to just stand there. so the AI would move them to the edge of the cover so it can scan for further danger.

other things in real life that give away a persons position is sound. could there be an area of awareness for every character (optional).

another fabulous feature would to have it control more than one object like say an infantry of characters.

climbing is another feature which I not to sure about how you would determine how to scan on a different level.

when an AI character reaches a cover position the eventonlocation would then set shikmesh animations into play.   

I think its important to get it right and feature packed with the correct A* algorithm. I am not sure Its necessary to have ballistics and other features which are already available.

n_iron

[archangel]

I think the AI should at least have a field of view that is definable in degrees and radius.... that way the AI could be initiated either by crossing into the FOV and/or entering the radius.  When the AI is initiated in pursuit,  it should maintain a radius from the target for melee and ranged attacks.  If the AI needs to flee,  it should just re-target to another object or vector.   If the obstacle avoidance feature is implemented, this should be sufficient.   Beyond that, maybe integrating the scanner object would be neat, but I don't know how plausible that is...

[Suna99]

If i knew ANY code at all i think i would be able to make atleast the cover function...(needs One Event on Location for each hiding spot+other EoL's for zones(places to hide from) and a path between them) Simple make one EoL a path and another EoL and link the Eol's with the path. Make it so when he wants to flee make the EoL's active. Do you get the point

EoL1=0 obstacle= | and EoL2=O->O|0
Path->                                      \/
Get it?

[Glitzfourjinn]

at shadmar list I would add a conditional:

if the player is looking at me, is more likely I will apply evassive tactics, but if is looking in other direction I will attack him. In order to do this I would apply the same principles: try to stay as far as possible from my partners, at the back at the player, far from walls, etc. but the point is: if the player is not looking at me the movement patern would be oriented to attack.

but if you have attack mode activated when the fov of player cant see enemy than enemy will get closest path to you and kill you? so have a random generated path with like the 2 post above me saying a scann thing to where they dont immediately try to kill player unless he meets certain conditions keeping the player safe and enemy smart

[Daniel Cremers]

Are we ever gonna get ai in 3drad?

[ulogix]

Not until there is an agreement as to the basic logic required, because many people will have different advanced AI requirements.

[Daniel Cremers]

 

[Daniel Cremers]

So how much longer? Can't we just get a basic ai and then just add new features later?

[ulogix]

So how much longer? Can't we just get a basic ai and then just add new features later?

What is the definition of basic AI?

[Daniel Cremers]

for example: move to player, avoiding objects. If in range stop and attack. If out of range stop chasing.

[ulogix]

for example: move to player, avoiding objects. If in range stop and attack. If out of range stop chasing.

OK, so in order to develop basic AI, you need to develop the specification for the basic AI which everyone agrees with. Only then would it be possible to implement it. And you would need to go into specific details in terms of the logic for the basic AI components.

For example, you specify avoiding objects. But how? Does the AI player walk towards the obstacle until within a preset distance and then walks around it? Or does it choose a direct straight path around it from it's current location? Which way round does it go? Clockwise or anti-clockwise? Does it run or walk? Does it slow down when emerging from behind the obstacle or does it simply run out? What happens if the AI player can't see you? What if the AI player is in an enclosed space and needs to get out via the door? At what point does the AI player stop heading towards the target? ie how far from the target before it stops and does something else like shoot?

Maybe it's posible to split AI capability into small sub-components where they can be linked to your AI player to build various levels of AI complexity.

For example, you could have these -

1. Search / detection of for target
2. Head towards target
3. Avoid obstacles within proximity
4. Route plan around obstacles from start location to target

Maybe you could have an AI poll to determine what AI requirements are common to most users and concentrate on those.

[Daniel Cremers]

I think all of that would be OK. It would just be a matter of choosing the cheapest option, so developemnt can finally start.
Errors with ai always will happen. But let's just have something to play with... 
I think splitting it up at beginning would be a bad idea. Beacuse it would take too long to get all features together at the end. Starting with a basic setup and then expanding would be OK.

[3D Rad Developer - Fernando]

The new Scanner object currently allows you to implement basic AI without coding. This AI would be based on scanning the surroundings of a character for obstacles (eg, using an EventTimer to start the Scanner object every second in front of a Character, to see if another character is there, and do something based on the result).

I could also try to estimate a cost for a new path-finding object, based on 'navigation meshes'. Path-finding is the basis of any Character AI, as far as I understand.