What are latest Researches in RTS Games AI
Hi all,
I'd like to know what are the latest researches related to RTS Games AI so I can add something new there (i.e., I want to hear something like: There's a problem in path finding). And I've read the post http://www.gamedev.net/community/forums/topic.asp?topic_id=500102 and got its points. Also if there are any posts related to How to develop an AI Engine for RTS Game kindly post it
Thanks in advance
Definitely a revamp of the control interface, for example support for multiple pointers (RTS realism while controlling a frontline with just one pointing finger? pleeease...), a.k.a. multitouch.
Not at all what the OP was talking about.
Notice he said AI. He even referred to path finding as an example of what he was talking about.
RTS AI is broken into two categories.
There is the PI-type and the non-PI-type.
PI = "perfect information"
Most RTS games use PI. In other words they allow the computer opponent to be aware of everything that is going on in the whole game state. They have no "fog of war" so to speak.
Which type of AI are you talking about? Then we can get more specific.
There are interesting challenges in both sets. non-PI is more interesting to me as it is an exercise in real game-state analysis and employs many brute-force-with-pruning methods/searches and has a long way to go before the industry will even have it in the "optimization phase." PI is only interesting so far in how to "play easy" in a manner that is most convincing. Basically how well can you give the other player a chance without letting him/her know you're peeking?
Notice he said AI. He even referred to path finding as an example of what he was talking about.
RTS AI is broken into two categories.
There is the PI-type and the non-PI-type.
PI = "perfect information"
Most RTS games use PI. In other words they allow the computer opponent to be aware of everything that is going on in the whole game state. They have no "fog of war" so to speak.
Which type of AI are you talking about? Then we can get more specific.
There are interesting challenges in both sets. non-PI is more interesting to me as it is an exercise in real game-state analysis and employs many brute-force-with-pruning methods/searches and has a long way to go before the industry will even have it in the "optimization phase." PI is only interesting so far in how to "play easy" in a manner that is most convincing. Basically how well can you give the other player a chance without letting him/her know you're peeking?
Thanks to all of you guys.
@ AlphaCoder:
That's exactly what I'm talking about. I'm interested in non-PI type so, what are the latest results in this area (My aim is to know a shortage and try to solve it)
@ AlphaCoder:
That's exactly what I'm talking about. I'm interested in non-PI type so, what are the latest results in this area (My aim is to know a shortage and try to solve it)
Didn't notice the AI keyword. Sorry.
Actually there's a significant amount of AI involved in UI design, e.g. context awareness in rubberband selection.
But if what you want is the classical understanding of AI (as in "brains"), then... in our project, we have a physically simulated solar system with individual actors using sensors to gain situational awareness (non-PI approach). The major disadvantage is _severe_ computational overhead, especially when dealing with visual sensors (render to texture, image pattern recognition and classification). And the major advantage is the system's modularity - we can for example grab the AI class controlling aircraft, optimize it, and upload it to an Arduino board controlling a small quad-copter, and surprisingly enough, it works as expected. We're working with a couple of space startup companies on orbital inertial navigation units hoping one day to hit space ourselves.
Actually there's a significant amount of AI involved in UI design, e.g. context awareness in rubberband selection.
But if what you want is the classical understanding of AI (as in "brains"), then... in our project, we have a physically simulated solar system with individual actors using sensors to gain situational awareness (non-PI approach). The major disadvantage is _severe_ computational overhead, especially when dealing with visual sensors (render to texture, image pattern recognition and classification). And the major advantage is the system's modularity - we can for example grab the AI class controlling aircraft, optimize it, and upload it to an Arduino board controlling a small quad-copter, and surprisingly enough, it works as expected. We're working with a couple of space startup companies on orbital inertial navigation units hoping one day to hit space ourselves.
One idea often discussed when RTS AI is brought up is influence maps. This assigns to each location on the terrain a number describing the extent to which it is "influenced" by some factor -- e.g., by the presence of enemy troops, or by that of your own troops (or the difference between the two). This can be visualized as a "density function." One can imagine influence maps for many ideas: Danger (due to enemy troops), resource availability, value as a chokepoint, etc. Influence from various sources of influence can, as appropriate, spread out in some radial manner (e.g., inverse square), or using a flood-fill algorithm.
Once one has influence maps, there are many things that can be done. For instance, one might find the center of mass of the opponent's influence, and then find minimum-enemy-influence paths from your base to theirs; this would correspond to attacking the weakest sections of their base...
Once one has influence maps, there are many things that can be done. For instance, one might find the center of mass of the opponent's influence, and then find minimum-enemy-influence paths from your base to theirs; this would correspond to attacking the weakest sections of their base...
I don't know if I would qualify influence maps as being the "latest research", however. It's old news from an algorithm standpoint and has been used in plenty of games.
While not an RTS, one place to get a very visual education of how influence maps are used is in Civ 4. The function that controls the societal influence is an influence map and is displayed in great detail right there in the UI.
While not an RTS, one place to get a very visual education of how influence maps are used is in Civ 4. The function that controls the societal influence is an influence map and is displayed in great detail right there in the UI.
Thanks guys. What about machine learning in RTS Games (i.e. discovery how player wins the computer and block him/her next time) is this area still open or it's closed?
It certainly is something that people are attempting to improve. It doesn't really make it into the commercial game market at this time, however.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement