Poll on Author awareness of Level Of Detail models in TDM

[i30817]

Read the last line of my last post (and the rest to see that i've noticed your previous objection too - thou maybe not to the point where i realized that would save nothing, but what they probably did).

To repeat it; why not do one more complete patrol route, since the last time the pc was 'near' a part of their route, and if nothing is discovered pause them?

I guess there can be artifacts if a player is moving in and out of parts of the route, so just checking pc 'near' the actor for pause/reactivation could lead to slight inconsistencies. But this sort of stuff could possible lead to deactivate a whole 'manor house' AI more or less seamlessly on a larger city. It would probably have to be the player object marking the npcs routes as being 'visited' thou, for obvious efficiency reasons.

 

edit: TBH, it might also be a case of 'optimizing the uncommon situation' since it's likely the player will bungle something, and that the AI will discover a body in said manor house, alert the others and never will come down from their alert state (they don't after murder right? And if you want this not to affect game mechanics, they shouldn't pause in this case... maybe). But at least at the start most AI in large levels wouldn't be thinking.

Edited March 1, 2013 by i30817

[jaxa]

How about AI that are an arbitrary distance away from the player proceed to the next node in their patrol/route and pause if the player is out of range. Once the player moves back into range they are activated again and they proceed to the next node. If you set the distance cutoff high enough then the player MIGHT not be able to leave a body lying around, run out of range, and exploit frozen AI. AI patrols are generally not synchronous but if desired several AIs could be linked together into a synchronized group, so the whole group continues to move if one of them is within range. To not affect existing maps you could set every AI to have a default distance cutoff of 9,999,999,999... etc. I admit that the balance between keeping AI responsive and improving performance might not work well in either direction at any point, though having 50-75% of AI in a map active could improve performance. Also, how about LOD AI behavior, where certain checks aren't performed at a distance, such as a missing loot check or a sound check (AI can alert AI, right?). You could draw a series of overlapping circles with the AI at the center to visualize this.

 

</brainstorm>

Edited March 1, 2013 by jaxa

[SeriousToni]

They had this in Thief 2 (and 1?) and this is a point where I like TDM much more, because it doesn't do this.

[Springheel]

_{Or more radical, just pause them}

 

You can already do this. Just set the AI to "neverdormant" "0", and they won't think or do anything else until they're in the same area as the player.

 

You can also increase the interleaved thinking value, so more frames go by between AI "thinking" (although this can cause problems above a certain level for patrolling AI).

[nbohr1more]

LOD on AI thinking would essentially boil down to reducing their path finding resolution at higher distances. The problem being whatever lower resolution representation you use has to have approximately the same chance of discovery as the full resolution representation. A complex problem. It has been surmounted by a few elite AI engineers commercially...

[nbohr1more]

One thing that might be nice is dynamic monsterclip or multiple precompiled AAS at different detail levels. That might be easier the rewriting the whole AAS system...

[grayman]

LOD on AI thinking would essentially boil down to reducing their path finding resolution at higher distances. The problem being whatever lower resolution representation you use has to have approximately the same chance of discovery as the full resolution representation. A complex problem. It has been surmounted by a few elite AI engineers commercially...

 

TDM already reduces pathfinding via "interleave thinking" which reduces the number of times an AI thinks when it's far from the player. Discovery isn't affected. Being able to handle doors and elevators isn't affected. Ever since The Alchemist's phantom killer was banished, we've been able to crank up amount of reduction to where an AI can skip as many as 32 frames between "thinks" w/o impacting his movement.

[grayman]

One thing that might be nice is dynamic monsterclip or multiple precompiled AAS at different detail levels. That might be easier the rewriting the whole AAS system...

 

I don't understand this. How would AAS at low detail (and what does that mean?) be different than AAS at high level? And what are the pros and cons of low and high AAS.

[nbohr1more]

I am running on the presumption that AI continue to search all reachable areas when out of sight. If those are setup to allow the player to get into all sorts of nooks and still be reachable by AI that is a lot of extra searching happening off screen. If AAS were simplified to just the AI route and trigger items that would save some processing. The problem with that approach is that players might stash bodies in those nooks so you would need the body itself to search for its own status and relay that to the AI