have unfair AI advantages been neutralized?

[Tinker]

Those are issues with all games, playing against the AI. GR1 is the same.

Tinker

You are wrong my Friend, there are games without these issues :

- Far cry

- Splinter cell(s)

- Chronicles of Riddick

- Call of Juarez

I Stand corrected. I`ve been running around a few other games, and i see were it`s going here now.

@ Cangaroo: Strange 1 about Vegas is, it depends were you are. Had the same as you in certain places, others they fired back, soon as i fired and took me down.

Anyhow, all AI Cheat. They Glitche, they even find hidden weapons in the middle of know were. And i swear i saw 1 using ret lock too.

Tinker

[Sleepdoc-iBeta]

I'm glad to see this issue take on such interest. I hope the good folks at GRiN took the time to upgrade the enemy (and friendly) AI in these ways.

The simple fact is that all these types of AI enhancements are possible in code. They just need to be designed for upfront, committmed to, coded and tested. Period. no magic here. I only need to play a little bit of FarCry to recall just how interesting AI can be.

Pretty much every aspect discussed here has been seen in one game or another. The trick is to get most of these behaviors and abilities (like blinded by smoke) into this game.

Here is to hopeful thinking.

Lets all return to this thread after relase, and discuss how all these AI elements turned out.

Edited May 25, 2007 by Sleepdoc

[Ron]

' date='May 25 2007, 10:47 AM' post='466834']

Thats odd... I've been shot a number of times by the AI in r6V through smoke after I open fire on them through the smoke.

I can tell you I did it last night. It was in the Calypso, in the small airlock between the two vault doors. I threw smoke in the airlock, killed the tango in the room, entered the room wherein tangos enter from the far door (must be an area trigger), went back into the smoke, and finished them off. Two were croutched, one was standing, but no one fired back.

Maybe I'm more intimidating

Oh don't get me wrong I've had those moments too but I dont see them offten... I think I am just a bullet spunge. lol

It's pretty funny on vent. to hear my buddys in my squad playing games like BF2/2142... there always saying things like "Why is it every time I spawn on you, you are always under heavy fire!" lol.

I Stand corrected. I`ve been running around a few other games, and i see were it`s going here now.

@ Cangaroo: Strange 1 about Vegas is, it depends were you are. Had the same as you in certain places, others they fired back, soon as i fired and took me down.

Anyhow, all AI Cheat. They Glitche, they even find hidden weapons in the middle of know were. And i swear i saw 1 using ret lock too.

Tinker

I think you’re dead on with that... I mostly play campus, Library, and research lab. We don't offten play maps like calypso because I noticed the map would take too long and other people would start to drop off before we could finish the map/round. In Library and Research Lab I do notice they take cover and fire on me more often then in other maps.

[Pritzl]

LoS implementations are actually quite tricky. If you're familiar with pathfinding algorithms, then about the same order of processing is required except (a) it's in 3D, (b) the routine cannot be exited the minute a "good-enough" path is found. In other words, I can see it being quite the processor hog to do it "right". Imagine e.g. every AI on the map running a ray-trace algorithm throughout the 3D sphere around him, constantly!

So, some shortcuts are taken. e.g. one interesting one for ray-tracing was to make it centre around the player rather than the AI. i.e. you track just the player's visibility to every point in his 3D sphere and if there's an AI within it, bam, you're spotted.

Some of the approximations/short-cuts taken in GRAW1 were admittedly a bit too lenient for my tastes and I hope that at least some of them were curtailed for GRAW2.

[Sleepdoc-iBeta]

e.g. one interesting one for ray-tracing was to make it centre around the player rather than the AI. i.e. you track just the player's visibility to every point in his 3D sphere and if there's an AI within it, bam, you're spotted.

I prefer ....

BAM .... you are now given 2 points toward being spotted where 10 points are required for fast spotting, 8 points for slow spotting, and 6 points for very slow spotting.

Now assign points once line of sight is established for the first 2 points.

A. If well lit where you are, add 4 points.

B. If in the shadows, add 2 points - add one more for excessive mvement in those shadows

C. If in pitch dark corner, add no point

D. If less than 300 feet from spotter, add 2 more points

E. If less tha 400 feet from spotter, add 4 points

F. If you make a low noise, add 2 points

G. If you make a loud noise, add 4 points.

etc etc.

You get the idea. In other words, LOS should only be one factor taken into account in "getting spotted". The more things you can model (Sound, lighting, movement, stance, etc) that add toward getting spotted in some logical fashion, the more motivated players become to be "ghost like" and stealthy. When every little action and nuance pays off, the more interesting each step you take becomes.

Edited May 27, 2007 by Sleepdoc

[Kung-Fu]

I remember a while back there was an Israeli company that was developing an AI processing card, much along the lines of the PhysX cards. Their goal was to release a toolkit that allows dynamic pathfinding and dynamic execution. They're intentionally flying under the media radar right now, but I think this would probably solve a lot of problems related to cheap heuristics. Hopefully they see the project through.

[Pritzl]

Actually Sleepdoc, that's why I likened it to path-finding. In the latter, various areas of the map have different "weights" depending on how crossable the terrain there is. e.g. impassable terrain might have a value of 1, rough 0.2 and paved roads 0. Then when you calculate the 'edges' as per A* you take these weights into consideration and any path that reaches a value of 1 is not considered. With LoS, you could then use weights as a measure of how hard it would be for light (rather than a unit) to cross it. This doesn't take into consideration things like dark to light and vice-versa of course though it might be doable via a vector representation of the edge weights.

In short, it's just a lot more complicated than many might think to do "right".

[Sleepdoc-iBeta]

Actually Sleepdoc, that's why I likened it to path-finding. In the latter, various areas of the map have different "weights" depending on how crossable the terrain there is. e.g. impassable terrain might have a value of 1, rough 0.2 and paved roads 0. Then when you calculate the 'edges' as per A* you take these weights into consideration and any path that reaches a value of 1 is not considered. With LoS, you could then use weights as a measure of how hard it would be for light (rather than a unit) to cross it. This doesn't take into consideration things like dark to light and vice-versa of course though it might be doable via a vector representation of the edge weights.

In short, it's just a lot more complicated than many might think to do "right".

I have no doubt it takes a lot of work and algorithm "tweaking" to get it right.

Edited June 2, 2007 by Sleepdoc