Well there was been some chat about group ais latelly.
I dont want to go exaclty into that line.
Rather want to ask somehow experienced players,
What role you think a Group Ai could fill that will improove YOUR gameplay?.
That is, related to your gameplay and the things you are bad at microing, or things you always forget to do.. or that are just too much time consuming that will justify using an ai for it.
To help you figure out what could this possibly be, think of what tasks you "delegate" when playing shared commander with someone else.
I for example, (if im in charge) always ask a player to mex all the place, (using 1+ cont units), so i can keep planning ahead.
or sometimes ask for as "scout" run on a specific zone, altougth i rather scout myself :).
e: gameplay*
About Group Ais
Moderator: Moderators
About Group Ais
Last edited by mongus on 23 Dec 2005, 22:24, edited 2 times in total.
Its a shame "experienced players" is in teh eye of teh beholder... some people that dont know what there on about are bound to make a suggestion that will be flamed, that will ruin this thread... which is a pity casue groupe AI's are a cool idea, but we need to be carfull how much of the game we remove.
aGorm
aGorm
I'd love an econ-building AI, but that's mostly because I can't effectively balance my attention between econ building and the rest of the game :)
What about having new buildings that enable different group AIs? You could have a lvl 2 constructor called the Programmer that builds them. They would need some (maybe a lot?) energy, and maybe the ai-controlled units would have to in radar range. These and other factors would eliminate the current all-or-nothing aspect of group AIs and add new strategic/tactical elements to the gameplay.
What about having new buildings that enable different group AIs? You could have a lvl 2 constructor called the Programmer that builds them. They would need some (maybe a lot?) energy, and maybe the ai-controlled units would have to in radar range. These and other factors would eliminate the current all-or-nothing aspect of group AIs and add new strategic/tactical elements to the gameplay.
-
Felix the Cat
- Posts: 2383
- Joined: 15 Jun 2005, 17:30
I think that GroupAIs should automate tasks that are either difficult or impossible for a human to do, or ones that are not significant in terms of strategy or tactics.
For example:
A GroupAI should keep your units in their formation and synchronize their move speeds to the slowest mover (as Simple Formation doesn't do). A GroupAI should not automatically take a group of units and raid with it.
A GroupAI should turn off your metal makers automatically when energy is low. A GroupAI should not determine when you need more metal makers or energy generaters and build them.
A GroupAI should automatically make units run away if fired upon. A GroupAI should not automatically make units determine if they should stand their ground, counterattack, or retreat if fired upon.
Thinking of TA as a war simulation: a GroupAI should do things that individual soldiers or very small units are capable of doing on their own, without input from above. A GroupAI should never manage things at the command scale - the player is the commander; the player has the sole ability to issue commands as the commander.
For example:
A GroupAI should keep your units in their formation and synchronize their move speeds to the slowest mover (as Simple Formation doesn't do). A GroupAI should not automatically take a group of units and raid with it.
A GroupAI should turn off your metal makers automatically when energy is low. A GroupAI should not determine when you need more metal makers or energy generaters and build them.
A GroupAI should automatically make units run away if fired upon. A GroupAI should not automatically make units determine if they should stand their ground, counterattack, or retreat if fired upon.
Thinking of TA as a war simulation: a GroupAI should do things that individual soldiers or very small units are capable of doing on their own, without input from above. A GroupAI should never manage things at the command scale - the player is the commander; the player has the sole ability to issue commands as the commander.
Actually you've hti the nail on the ehad witht eh perfect GroupAI solution there.
Forcing a GroupAI to search for a building so say the groupAI wont do anything if you havent built building x or control centre x or mobile unit b, or you can assign a decoy commander as a group leader but that group can only handle 5 other con untis after whcih you need another decoy commander. I think that that si the only valid way of getting these types fo AI's in without ti simply being, "hey, here's a file that'll play this bit fo the game for you!" because there is strategy involved in getting that ability and it becomes a part of the modders domain and not simply of freewill.
If any modder out there is interested in testing this out to see how effective it is I wouldnt mind co-operating at all.
Forcing a GroupAI to search for a building so say the groupAI wont do anything if you havent built building x or control centre x or mobile unit b, or you can assign a decoy commander as a group leader but that group can only handle 5 other con untis after whcih you need another decoy commander. I think that that si the only valid way of getting these types fo AI's in without ti simply being, "hey, here's a file that'll play this bit fo the game for you!" because there is strategy involved in getting that ability and it becomes a part of the modders domain and not simply of freewill.
If any modder out there is interested in testing this out to see how effective it is I wouldnt mind co-operating at all.
I don't think so. In the case of the mex-spammer, requiring a decoy commander to lead the construction units means you've got to aquire level 2 kbots before you can use it. However, it's mainly during the early game that such a group AI would be useful, where one needs to expand quickly while at the same time being preparing to deal with a quick rush (or launching one yourself). By the time you get to level 2 you've either established a good resource system already or you're so far behind that you're in dire trouble.Alantai Firestar wrote:Actually you've hti the nail on the ehad witht eh perfect GroupAI solution there.
Besides, these things are war machines from over 4000 years in the future with cloned or patterned human brains running them. Why can't the "subcommander units" for these group AIs be the construction units themselves? They aren't morons, they should be able to handle vague instructions like this.
Somehow I think a human mind cloned thousands of times over many times epr second for thousands fo years would suffer some degradation at least.
And I wasnt thinking that a mex spamming GrouPAI would require a lvl 2 decoy commander, I was just usign a decoy commander as an example, if it where for a lvl 1 type tactic then a lvl 1 type building would need to be added, perhaps a control tower of sorts with a bit of radar on it.
Overrall I think this si somethign the modder would ahev to deal with so we neednt worry ourselves about having decoy commander for this etc because we're the player not the modder, and the modder will know what theyw ant and how this best fits in or if it should fit in at all.
And I wasnt thinking that a mex spamming GrouPAI would require a lvl 2 decoy commander, I was just usign a decoy commander as an example, if it where for a lvl 1 type tactic then a lvl 1 type building would need to be added, perhaps a control tower of sorts with a bit of radar on it.
Overrall I think this si somethign the modder would ahev to deal with so we neednt worry ourselves about having decoy commander for this etc because we're the player not the modder, and the modder will know what theyw ant and how this best fits in or if it should fit in at all.
Something that I think would be VERY useful, would be patrol replenishment after casualties.
To explain:
Ok, there's a sensitive area you want patrolled by your units, wether it's a scout mission or an area denial mission. You set a group of units to patrol that region.
The enemy then crosses your patrol, a fight takes place, and your patrol takes 50% casualties. Now, on top of everything else you're doing, you need to build replacements, reselect the group, and reassign the patrol route.
If that patrol was given an AI group for auto replenishment, then they would send a command to a factory or to currently idle standby units to fill in their ranks.
There's tons of possible issues with this idea, but it's still a good idea, as that's one of the major distractions during gameplay.
Another idea, similar, is auto replenishment of defensive structures. If you set a defense / building to this, once it's destroyed, an idle construction unit will go there and start building a new one. Or rez bots.
To explain:
Ok, there's a sensitive area you want patrolled by your units, wether it's a scout mission or an area denial mission. You set a group of units to patrol that region.
The enemy then crosses your patrol, a fight takes place, and your patrol takes 50% casualties. Now, on top of everything else you're doing, you need to build replacements, reselect the group, and reassign the patrol route.
If that patrol was given an AI group for auto replenishment, then they would send a command to a factory or to currently idle standby units to fill in their ranks.
There's tons of possible issues with this idea, but it's still a good idea, as that's one of the major distractions during gameplay.
Another idea, similar, is auto replenishment of defensive structures. If you set a defense / building to this, once it's destroyed, an idle construction unit will go there and start building a new one. Or rez bots.
Both of whcih are already possible using sub-unit behaviours and tactics.
Set a construction unit to build defences but with repeat orders turned on. If one of the structures is blown up, it rebuilds them. Works for any cosntruction be it a set of units, missile towers, geothermals factories dragonteeth etc....
Tell your factories to make units they build patrol on these areas. When you need to send otu units take them from areas on patrol, and they'll be replenished by the factories, and if you are busy with toher things your patrols will get bigger and bigger as big roving defences ready to attack when you get back to them.
All done without any GroupAI's at all.
What you want can be done already either with sub unit AI's, the repeat command, or the construction AI. Set a mass of con units under this AI then place a mex on all the spots quickly, the con AI will sort out whcih goes where and does what for you.
When that con is captured/destroyed they blame the AI rather than realize it was their own foolish mistake, and demeaning tot hem to expect the AI to be capable of greater abstract reasoning than them even though they know it cannot be done.
Totally invisible bases are already possible in sprign and have been in OTA since it was first brought out in 1997, the only difference is that modders know that if a player can cloak there base to make it totally invisible they will do so and games will never be won. Any cloaked unit under jammer radius is totally invisible to the enemy untill one or the other is removed.
*and I did see that sarcasm but I felt it wasnt enough to quash the idea*
The worst thing you could do at that point is rely on an economy groupAI.
People like storm and so on didnt realize a simple truth that we did, but they feared this but for different reasons, reasons whcih while valid are not the most probable outcome.
I see if these uber AI's are made and released people will rely ont hema nd the game will become a 3 tier class community.
The n00bs who rely on groupAI's rather than learn how to handle their game correctly, who end up putting more effort itno managing to work with the AI than would ebr equired to learn tit hemselves and get better results.
The newcomers who come from other games and are willing to learn but face problems that there is a choice between climbing the hill to beat the n00bs then being propelled quickly into the 3rd class I'll mention shortly, or joinign the n00bs using groupAI's.
And thirdly the pros, the people who refuse to use them, or simply find them innefficient. They'll amssacre the first group easily through superior skill and a better grasp of the game. Some may even know how to elverage a groupAI to their advantage in a way that doesnt mean they rely totally on them, jsut as a passing tactic, but mostly they'll be the sort who are at the top of the community int erms of ability.
GroupAI supporters are sometimes suggesting groupAI's just for the sake of it, or dont realize that it's alreayd possible using a few clicks using engine features and a bit of skill, withotu requiring masses of micromanagement at all. Some of you haven't even looked at all the buttons in the side panel.
GroupAI's are there to bring new features to the game, or to automate tasks that are totally annoying such as keepign things in formation, or turnign metal makers on and off, things that a human can do but just barely and not very well. Not automating things a human can do, be it with relatvie ease or with skill, such as placing mexes or dealign with patrols.
Besides the things you suggest are reliant on things that humans excel at and are not numerical in nature. For example how does the AI know when tor eplenish a patrol? Yah you've taken 50% out but what was the figure you anted in the first palce? Maybe when you asigned it it had 50 in patrol, but where those 50 a full patrol or did you assign it because you saw that you'd taken a load and 50 units where left? What units should it replace them with? Was there a specific ratio of types? And what should it do with the factory when it's replenished the patrol? Should it build cons or attack units if so which ones and how does it know what you want? More to the point when should it change the patrols? Or realize that it has to give up as there's now say it can replenish and if so how does it tellt eh user?.
Sorry for the huge post but this whole thing is getting annoying now, and very n00bish, and it's icnreasingly hard to keep in the middle of both sides of for and against.
Set a construction unit to build defences but with repeat orders turned on. If one of the structures is blown up, it rebuilds them. Works for any cosntruction be it a set of units, missile towers, geothermals factories dragonteeth etc....
Tell your factories to make units they build patrol on these areas. When you need to send otu units take them from areas on patrol, and they'll be replenished by the factories, and if you are busy with toher things your patrols will get bigger and bigger as big roving defences ready to attack when you get back to them.
All done without any GroupAI's at all.
What you want can be done already either with sub unit AI's, the repeat command, or the construction AI. Set a mass of con units under this AI then place a mex on all the spots quickly, the con AI will sort out whcih goes where and does what for you.
This is a vital part of gameplay, one that a good player will use to there advantage, it isnt always just about filling gaps. Besides an AI would always end up building them in a predictable place and it'd be a lot easier to get rid of the enemies radar coverage if they used this or even to capture cons by sending capture units to places waiting for the lone con to walk out under AI control, even tho the player is totally oblivious 'because the AI is taking care of it'.I guess I'd like to see an AI that adds radar and/or jamming wherever you have gaps. Well, maybe not a whole lot, but someone else might.
When that con is captured/destroyed they blame the AI rather than realize it was their own foolish mistake, and demeaning tot hem to expect the AI to be capable of greater abstract reasoning than them even though they know it cannot be done.
Totally invisible bases are already possible in sprign and have been in OTA since it was first brought out in 1997, the only difference is that modders know that if a player can cloak there base to make it totally invisible they will do so and games will never be won. Any cloaked unit under jammer radius is totally invisible to the enemy untill one or the other is removed.
*and I did see that sarcasm but I felt it wasnt enough to quash the idea*
That is not somethign groupAI should be dealign with, that is either micromanagement, or a sub-ai behaviour that needs changing. I'm more inclined to the latter.A GroupAI should automatically make units run away if fired upon. A GroupAI should not automatically make units determine if they should stand their ground, counterattack, or retreat if fired upon.
This is a good suggestion and would take a minute to implement.A GroupAI should keep your units in their formation and synchronize their move speeds to the slowest mover (as Simple Formation doesn't do). A GroupAI should not automatically take a group of units and raid with it.
If you cant handle your economy, then an AI is only going to make you worse when you rely on it. Balancing your economy is itnegral to how you handle your forces. If you make an AI run your economy for you then you have to follow a certain strategy to win the game that follows what your AI partner is doing. How you attack and defend directs your economy and vice versa, you cannot seperate them and expect better results even if you're not good at doing one of them anyway.I'd love an econ-building AI, but that's mostly because I can't effectively balance my attention between econ building and the rest of the game :)
The worst thing you could do at that point is rely on an economy groupAI.
People like storm and so on didnt realize a simple truth that we did, but they feared this but for different reasons, reasons whcih while valid are not the most probable outcome.
I see if these uber AI's are made and released people will rely ont hema nd the game will become a 3 tier class community.
The n00bs who rely on groupAI's rather than learn how to handle their game correctly, who end up putting more effort itno managing to work with the AI than would ebr equired to learn tit hemselves and get better results.
The newcomers who come from other games and are willing to learn but face problems that there is a choice between climbing the hill to beat the n00bs then being propelled quickly into the 3rd class I'll mention shortly, or joinign the n00bs using groupAI's.
And thirdly the pros, the people who refuse to use them, or simply find them innefficient. They'll amssacre the first group easily through superior skill and a better grasp of the game. Some may even know how to elverage a groupAI to their advantage in a way that doesnt mean they rely totally on them, jsut as a passing tactic, but mostly they'll be the sort who are at the top of the community int erms of ability.
GroupAI supporters are sometimes suggesting groupAI's just for the sake of it, or dont realize that it's alreayd possible using a few clicks using engine features and a bit of skill, withotu requiring masses of micromanagement at all. Some of you haven't even looked at all the buttons in the side panel.
GroupAI's are there to bring new features to the game, or to automate tasks that are totally annoying such as keepign things in formation, or turnign metal makers on and off, things that a human can do but just barely and not very well. Not automating things a human can do, be it with relatvie ease or with skill, such as placing mexes or dealign with patrols.
Besides the things you suggest are reliant on things that humans excel at and are not numerical in nature. For example how does the AI know when tor eplenish a patrol? Yah you've taken 50% out but what was the figure you anted in the first palce? Maybe when you asigned it it had 50 in patrol, but where those 50 a full patrol or did you assign it because you saw that you'd taken a load and 50 units where left? What units should it replace them with? Was there a specific ratio of types? And what should it do with the factory when it's replenished the patrol? Should it build cons or attack units if so which ones and how does it know what you want? More to the point when should it change the patrols? Or realize that it has to give up as there's now say it can replenish and if so how does it tellt eh user?.
Sorry for the huge post but this whole thing is getting annoying now, and very n00bish, and it's icnreasingly hard to keep in the middle of both sides of for and against.
I think there's a command for that. Someting like holding ctrl, or alt, while giving a move order. What is Simple Formation and what does it do btw?Felix the Cat wrote:A GroupAI should keep your units in their formation and synchronize their move speeds to the slowest mover (as Simple Formation doesn't do).
As far as I played Spring, this Group AI was already there.Felix the Cat wrote:A GroupAI should turn off your metal makers automatically when energy is low.
Ok, but only if this is disabled by default. I played a RTS where units thought they were being by retreating while under fire, and it was a disaster, when I told them to attack long ranged turret, they moved back forth just inside enemy range and just outside their range till they were dead, and wouldn't want to move forward under fire no matter I clicked. Please spare me that experience in Spring.Felix the Cat wrote:A GroupAI should automatically make units run away if fired upon.
Like AF, said, make sure your construction units have repeat ON when you give them the order to build the defense, and they will do that.tanelorn wrote:SAnother idea, similar, is auto replenishment of defensive structures. If you set a defense / building to this, once it's destroyed, an idle construction unit will go there and start building a new one. Or rez bots.
IMO, group AI maker should just go make group AI and propose them and we'll then discuss on the forum how well their perform, because I'm convinced that for group AI the usefullness or uselessness is much more related to the details and to how it is coded than to the general idea of what it is supposed to do. And contrary to others, I don't think AI will ever be as good as to completly replace human in Spring games. And if they ever do, well, then, and only then, we'll have to add something to the lobby to let the host restrict group AI.
I think a group AI that allows you to define complex formation behavior would be good;
Example: Penetrators can fire over a bulldog- a group AI should exist which when properly put together by the user will cause a penetrator to 'hide' behind the nearest bulldog (behind being with respect of where the last move/attack statment was given, or perhaps you identify direction when you build the group)
Example2:Crawling bombs expode with radius 23 (made up number, I have no idea how big a crawling bomb explodes with)- there should be a formation AI that would allow you to specify that any crawling bombs in that group stay 24 apart from other units (including other crawling bombs).
Example3: Shooters have range 70, and bulldogs have range 40 (again, made up numbers); it should be possible to tell shooters to stay 30 back from the bulldogs, but make sure they have line of sight for firing to the last location clicked.
This should be a single AI that calls configuation files that the user has built- so that it gives good extensibility to the user. It should serve to make groups of mixed units behave as a single effective squad, rather then a pile of units that will require microing to make effective.
Example: Penetrators can fire over a bulldog- a group AI should exist which when properly put together by the user will cause a penetrator to 'hide' behind the nearest bulldog (behind being with respect of where the last move/attack statment was given, or perhaps you identify direction when you build the group)
Example2:Crawling bombs expode with radius 23 (made up number, I have no idea how big a crawling bomb explodes with)- there should be a formation AI that would allow you to specify that any crawling bombs in that group stay 24 apart from other units (including other crawling bombs).
Example3: Shooters have range 70, and bulldogs have range 40 (again, made up numbers); it should be possible to tell shooters to stay 30 back from the bulldogs, but make sure they have line of sight for firing to the last location clicked.
This should be a single AI that calls configuation files that the user has built- so that it gives good extensibility to the user. It should serve to make groups of mixed units behave as a single effective squad, rather then a pile of units that will require microing to make effective.
-
PauloMorfeo
- Posts: 2004
- Joined: 15 Dec 2004, 20:53
Re: About Group Ais
So much to talk about, so little time...mongus wrote:... about group ais...
What role you think a Group Ai could fill that will improove YOUR gameplay?.
...
One of the things i most have problems with is waiting times. For example:
- Building moho mines. Usually, in it's place there is a metal extractor. Many times it is usefull to reclaim it to get back the 50 Metal it costed. You have to either wait for the builders to reclaim it (losing precious time) or leave the builders with the reclaim orders and go away do something else. This leaves unfinished businesses in my mind (using precious RAM) and i often forget about it, coming back much later.
Ok, it's only 50 Metal, just blow it up. In XTA they're even ready for that with a zero cout-down time of self destruction so we don't have to wait. But i want to build more than one moho. I want to build 2~3~10 mohos. So what do i do? Leave one building and go away, leaving "unfinished businesses in my mind (using precious RAM) and i often forget about it, coming back much later", or just self-D 10 mexxes? That isn't just 500 metal, it means that, until they're replaced with mohos, they are not producing any metal crushing your metal income.
To this specific case, i can't think of any solution other than making the mexxes be auto-reclaimable like Dragon Teeths. (it isn't really a Group AI solution unless a GAI work around can be made?)
Still, there are many other situations where we have to leave pending issues. That has a big problem with scalability because it isn't a problem when we have a small army having 1~3 pending issues but when the armies/bases start to scale up, so does the pending issues...
-
PauloMorfeo
- Posts: 2004
- Joined: 15 Dec 2004, 20:53
One more other thing, i really wish units could be ordered to auto-scouting/auto-harassment roles.
(like TA's «brother» Dark Reign and many others have)
That way the game would scale very well because we could almost as easily handle 2 scouts or 20 harassers. (even though a human controled scout/harasser is much more usefull)
Think of a map like Greenhaven or Plains and Passes (which i don't play well at all). In them, a player that has monstruous micro management capabilities will have a huge advantage over one that doesn't. The mexxes are very scatered and low productive so you can't protect them all. That means that mexx raiding is extremely productive. So we end up starting with having to micro manage some 1~3 raiders at first, escalading to the need to micro manage, literally, dozens of scatered raiders.
(like TA's «brother» Dark Reign and many others have)
That way the game would scale very well because we could almost as easily handle 2 scouts or 20 harassers. (even though a human controled scout/harasser is much more usefull)
Think of a map like Greenhaven or Plains and Passes (which i don't play well at all). In them, a player that has monstruous micro management capabilities will have a huge advantage over one that doesn't. The mexxes are very scatered and low productive so you can't protect them all. That means that mexx raiding is extremely productive. So we end up starting with having to micro manage some 1~3 raiders at first, escalading to the need to micro manage, literally, dozens of scatered raiders.
[random idea]
I, like Paulo get annoyed with upgrading mexes. What would be handy (and quite possible) is for mexes to have the ability to 'upgrade' to Moho mexes. You just click a button in the mexes menu, and the mex will start constructing a moho mex on top of itself. Once done, the mex would quitely destroy itself. No more problems.
I, like Paulo get annoyed with upgrading mexes. What would be handy (and quite possible) is for mexes to have the ability to 'upgrade' to Moho mexes. You just click a button in the mexes menu, and the mex will start constructing a moho mex on top of itself. Once done, the mex would quitely destroy itself. No more problems.