Gaming AI

[GGTharos]

A lot of you are wondering about AI doing this and that in games, and perhaps how it all works.

 

Here's a relatively interesting article - or so I think - on AI:

 

http://www.bit-tech.net/gaming/2009/03/05/how-ai-in-games-works/1

[159th_Viper]

Interesting Indeed :)

 

Begs the question - Why is the AI in LockOn/DCS so very far behind the curve?

 

Engine limitations?

[GGTharos]

I can only guess, but there's a couple reasons:

 

One is as you said, engine limitations - the AI is being rewritten and updated, so I think you're seeing a mix of old and reasonable simplistic AI mixed with new and potentially somewhat incomplete AI behavior.

 

The other is environment processing limitations. ED can't just tag the whole world - it's not feasible, so the AI actually has to be a lot smarter about sensing its own environment - this is actually quite a difficult proposition that scientists have been working on for a while.

 

Without sensing the environment adequately, you can't make good decisions.

 

And also, there's the issue of priorities. While ED could develop a robust AI system, they can't focus -just- on that.

[159th_Viper]

 

And also, there's the issue of priorities. While ED could develop a robust AI system, they can't focus -just- on that.

 

Aye - personal ponder is that that pretty much sums it up!

 

And whilst putting the 'Intelligence' back into 'AI' is pretty high on my personal wish-list for the DCS series, it's akin to the Cream and Castor Sugar on the Strawberry - I'm pretty happy with the Berry on it's own atm......the rest will inevitably follow as is the nature with all things technological.

 

Interesting Indeed - one can but glimpse at the amount of work involved that one takes for granted.........:)

[EagleEye]

Yes, very interesting.

So I understand why it must be so complicated or just impossible to make solid clouds AI can`t look through.

I would like to see a article about AI with examples from simulation games.

[glottis]

Good summary article - I've always been really interested in Jeff Orkin's work with the Goal-Oriented planners in FEAR and No One Lives Forever 2. I'm just getting stuck into Empire:Total War where they've apparently used a similar system this time around, so I'll have to see if it's any good.

[RedTiger]

Is anyone here familiar with Battle of Britain II?:

 

http://www.a2asimulations.com/bob/

 

People rave about the AI in that sim being very human-like. On the rookie settings the pilots are more timid and easier to sneak up on. On advanced levels, they are very aggressive. On all settings however, they remain "human". They don't have super situational awareness. They are blinded by the sun and cannot see through parts of their plane that would block the pilots vision. You can actually sneak up on them or bounce them from above.

 

I've always wondered how that AI was created and if something like that could be applied to DCS.

[GGTharos]

DCS already has the LOS limitation - as for 'human', well ... that's another ball of wax.

[RedTiger]

DCS already has the LOS limitation - as for 'human', well ... that's another ball of wax.

 

I own BoB II but unfortunately have never found the time to play it much. Beyond a quick 1 vs 1 dogfight with the AI, I can't speak from experience. What people say about the AI in BoB II is that the flaws in the AI are more like they are built-in to make them more human rather than just being flaws in crappy AI. They make mistakes, but ones that make sense for a human to make. I know that BoB II's AI is probably proprietary, but you have a flight sim with no multiplayer, game play propped up by nothing but AI that some people really seem to enjoy. Yet other than its fans, no one pays any lip service to it. Why aren't more trying to copy it?

 

I have felt for a long time that there's a tendency to not give a crap about AI and just let competitive multiplayer provide all the replay value and longevity. I fully understand why a great deal of people prefer it that way, but there's just only so much that can be done when every unit has to be player-controlled. I'm not even talking about the scale (although that's a big part of it too) and number of units, its also practicality. How many players would want to sit in the Zulu barn for hours and hours for a scramble that may never come? This is where good AI and scripting has its strong points. The AI doesn't care if my flight never satisfies the conditions to trip the trigger to spawn an interceptor scramble. And if it does intercept me, it would be great if it weren't totally predictable and easily dispatched.

Edited March 7, 2009 by RedTiger

[Grimes]

Ahh AI in games. I find it fascinating but is completely out of my scope of comprehension, unless the game engine has a great editor. I remember reading about a technology John Carmack developed known as "Megatextures" it basically uses an extra normal map texture to define terrain. In the game Quake Wars, movement behavior was changed based on the type of terrain they were on. On a road a wheeled vehicle could have a higher top speed, when you went off roading the driving characteristics changed dramatically. I believe it used a form of normal maps (super cool advanced bump map textures) to define the surface.

 

Ideally though for a flight sim you need a thinking AI. One that learns where they are being attack from and move to a safer location to lower the probability of being destroyed. AI need to have objectives, either targets or self-preservation and to react accordingly.

 

But thats out of the question. I honestly believe the AI in future DCS games doesn't need to be so advanced that they dynamically react to everything. Rather have set predeterimined behaviors for it to fall back on under specific conditions. Something as simple as having an armor column turn around when <15% of their original forces remain, or going into a defensive tactical formation once they come under attack or realize that hostile air units are in the area.

 

Oops almost turned this into a AI wishlist thread ^_^