Space Combat (madness)

For what's not in 'Top Priority Game Design'. Post your ideas, visions, suggestions for the game, rules, modifications, etc.

Moderators: Oberlus, Oberlus

Message
Author
User avatar
utilae
Cosmic Dragon
Posts: 2175
Joined: Fri Jun 27, 2003 12:37 am
Location: Auckland, New Zealand

#151 Post by utilae » Wed Oct 13, 2004 6:55 am

Ranos wrote: It is a perfect union between Real-time and Turn based combat.
Oh yes, it is. Where did the idea of phased time combat come from anyway, other than my space combat proposal. I don't know if it has been used in other games or if I thought of it (lol I was one of the first to mention it for FreeOrion, probably the first, but surely it has existed before).
Ranos wrote: In hot seat, the battle begins and player 1 gives his orders. When player 1 is done, player 2 gives his orders. Then they get to watch their orders played out.
You could even have more than two players. Another plus, I think the fact the phased time allows more than three players to give orders at once (not in hotseat, in hotseat each player would have to give orders after the other player).

emrys
Creative Contributor
Posts: 226
Joined: Fri Oct 24, 2003 3:44 pm

#152 Post by emrys » Thu Oct 14, 2004 4:20 pm

RE- haravikk's ideas of 'bullet-time' and 'Order-requesting-units'.
MisterMerf wrote: I think the devil in those details will be considerably nastier than other AI devils. If ship AI can't carry out missions perfectly, as least it carries them out predictably.

<snip>

Even the pace of combat might be intelligently controllable by AI.

<snip>

haravikk seemed to be suggesting that the AI could decide when the player needed to give orders.
Much rests on the interpretation of 'needed to' here I feel.

I would agree that the AI should definitely NOT be relied upon to decide when a player should be ALLOWED to give orders, which would indeed be guaranteed to annoy everyone. I was thinking instead along the lines that players would be able to give orders at whatever time they would otherwise have been allowed (i.e. in the order phases for phased combat, any time for real time), but the AI would flag certain ships/task forces as needing orders. The player would be free to give orders to ships not flagged, and to not give orders to those that were flagged, who would then just do the best they could. The flags would give the AI a way of saying "I'm out of my depth here." rather than just leaving it to do dumb things without the player realising there was a problem.

The point was that I see haravikk's ideas as a way to turn two of the big problems with AI control/assistance of combat into advantages.

1) AI routines tend to be unable to cope when combat get's frenetic, since working out what to do rapidly becomes very complicated and starts to take more time than you have available. Slowing down the rate of time flow here would give the AI more time to 'think', whilst looking like a cool feature rather than just the game lagging.

2) AIs work fine at automating player choices when they know what to do, or the situation is simple, but sometimes start doing stupid things or cycling between actions because they can't see a way to make a clear decision. If we explicitly flag that the AI is uable to cope here, we turn this failing into a gameplay element. We no longer have to write an AI that can cope with everything, the player feels they are making the big decisions, and yet the player doesn't HAVE to handle the trivial details, because those the AI can manage without help.

User avatar
MisterMerf
Space Squid
Posts: 67
Joined: Sat Oct 02, 2004 3:38 am
Location: Saint Paul, MN (USA)

#153 Post by MisterMerf » Thu Oct 14, 2004 5:11 pm

emrys wrote:The player would be free to give orders to ships not flagged, and to not give orders to those that were flagged, who would then just do the best they could. The flags would give the AI a way of saying "I'm out of my depth here." rather than just leaving it to do dumb things without the player realising there was a problem.
That is a truly excellent idea.
We just need to make sure the programming team is aware of or agrees with it. I think having an AI that knows when it is confused is a handy thing, but will be difficult to add in after a working AI alredy exists.

/me prepares to spam the Programming forum ^_-

discord
Space Kraken
Posts: 199
Joined: Fri Jul 25, 2003 12:00 am

#154 Post by discord » Thu Oct 14, 2004 6:14 pm

emry's/merf, bad idea.

why? because if the AI is smart enough to know when something is wrong, it would be a rather short step to making it smart enough to do something about it, or in other words to damn *bleeeping* complicated.

//discord

Ranos
Dyson Forest
Posts: 234
Joined: Tue Mar 23, 2004 6:24 am
Location: Northern Wisconsin

#155 Post by Ranos » Thu Oct 14, 2004 7:15 pm

Did you ever play MOO3 discord? Now hopefully our guys can do a better job of programming the AI than QSI did, but if they can't or that is the best an AI can be with our technology today, then oh well.

If you never played MOO3, the AI would have this annoying habit of seeming to avoid the enemy TF, when they were a direct fire TF and had been ordered to attack the enemy. The enemy would be on one side of the planet and instead of the AI moving to intercept it, it would start flying around the opposite side of the planet. Now this was probably something that is programmed in so that the two TFs wouldn't get extremely close to eachother, but it was very aggrivating. Then the TFs would, for no apparent reason, twirl a circle or two out in the middle of nowhere instead of advancing towards the enemy.

Those are just a couple of examples of the AI getting confused. Now I don't know how to program and I don't know what the AI in MOO3 was told to do, but it seems to me like it shouldn't be that complicated to tell the AI to go from point A (its start location) to point B (the enemy TF) and attack. Now anyone who is old enough to program should know that the shortest distance between two points is a line. Not a line with a couple of twirllies and not a curved line around the far side of the planet, but a line heading straight towards the enemy TF.

I think the idea is a good idea, if it is needed. Maybe the AI will be smart enough to handle these kinds of issues. Only time will tell.

User avatar
haravikk
Space Kraken
Posts: 104
Joined: Wed Mar 31, 2004 10:04 pm

#156 Post by haravikk » Fri Oct 15, 2004 9:23 am

discord wrote:if the AI is smart enough to know when something is wrong, it would be a rather short step to making it smart enough to do something about it
Not really. For example, if a task-force completes its original mission, then the AI would flag this up before doing some default action.
However, the default action (which you seem to be suggesting would be chosen rather than prompting for a decision) could be to locate/attack the nearest enemy. The player might not want this, and that would be frustrating AI.
Say you have a short-range attack fleet, you order it to close on a planet and bombard it to destroy its defences, the default action once this has happened could be to find nearby enemies, however the player might have intended for that task-force to go onto he next planet and destroy its defences too. To solve that you'd have to get the AI to automatically try and repeat it's previous orders before it goes onto default ones, but then you might end up with the task-force dying against a superior planet the player wouldn't want to have attacked. So now you have the AI trying to do it's previous mission, but only if there is a planet it thinks it can beat, but this planet might be far away and of little immediate consequence.
And so you have ever increasingly more complicated AI to automatically do something which might not be what the player wants anyway, players are after-all very odd people sometimes and might try something that wasn't considered at all for the AI. For example using a task-force to draw enemy fire from a more powerful one.

But instead you could just have the task-force commander go into a sort of 'safe' decision mode and ask for a solution while doing their best to still be around for when they receive one.

Ranos
Dyson Forest
Posts: 234
Joined: Tue Mar 23, 2004 6:24 am
Location: Northern Wisconsin

#157 Post by Ranos » Fri Oct 15, 2004 2:50 pm

@ haravikk

I'm pretty sure someone said this earlier but I'll say it again. Your idea is nothing more than a complicated version of real-time combat. If I had to choose between normal real-time where I give the orders when I want to and your version where I can only give orders when the AI asks me to, I would choose normal real-time.

One of the problems with yours is there is now way for the AI to take into account other things going on in the battle. If I have three TFs and I send one to attack a planet and the other two to attack enemy TFs, they will go and do those jobs to completion. What happens if one of my TFs attacking the nemy TFs is destroyed? Lets say it was 2 on 2. Although my TFs did damage to one or both of the enemy TFs, it is still 2 on 1 now and I have a greater chance of losing that one TF. I want to send my TF attacking the planet over to help my other TF. With noraml RT, I could just order it over there. With your version, I would have to wait.

Now maybe the AI could be programmed to take all of this into consideration. That could take a lot more programming and a lot more of a hassle for our programmers.
200 and still a Wyrm!?! I don't want to be a Wyrm anymore. I've been a Wyrm for 100 posts now.

User avatar
utilae
Cosmic Dragon
Posts: 2175
Joined: Fri Jun 27, 2003 12:37 am
Location: Auckland, New Zealand

#158 Post by utilae » Fri Oct 15, 2004 9:57 pm

Ranos wrote: Now maybe the AI could be programmed to take all of this into consideration. That could take a lot more programming and a lot more of a hassle for our programmers.
And the AI takes a lot of CPU power, so if we have too much AI it would likely be bad. Plus like Drek says: "there's always a better alternative to AI control systems and to sliders" (something like that, lol).

How many games have actually had AI control systems that worked 100% to the players liking. I bet there are few, if any that work very well (Moo3 is not at the top of the list of good AI Control systems).

User avatar
Daveybaby
Small Juggernaut
Posts: 724
Joined: Mon Sep 01, 2003 11:07 am
Location: Hastings, UK

#159 Post by Daveybaby » Fri Oct 15, 2004 11:20 pm

The problem with getting a good AI isnt the necessarily amount of computing power it requires (to simulate the way a human player makes decisions is still several orders of magniture beyond our capability anyway), its figuring out how to program the damn things.
The COW Project : You have a spy in your midst.

User avatar
MisterMerf
Space Squid
Posts: 67
Joined: Sat Oct 02, 2004 3:38 am
Location: Saint Paul, MN (USA)

#160 Post by MisterMerf » Sat Oct 16, 2004 2:26 am

Daveybaby wrote:The problem with getting a good AI isnt the necessarily amount of computing power it requires (to simulate the way a human player makes decisions is still several orders of magniture beyond our capability anyway), its figuring out how to program the damn things.
???

In some cases, those two things will be inextricably linked. For instance, let's say I can give my AI an excellent state to work from if only I run an algorithm that bears a little too much resemblance to the knapsack problem. Now all of a sudden, I'm limited to 15 task forces per battle before it starts taking years to update the AI's conception of the battle.

In choosing AI strategies, there will also be a lot of algorithm design that would surprise me if some of it didn't have to choose between "intelligence" and performance.

Ranos
Dyson Forest
Posts: 234
Joined: Tue Mar 23, 2004 6:24 am
Location: Northern Wisconsin

#161 Post by Ranos » Sat Oct 16, 2004 4:34 am

utilae wrote:
Ranos wrote: Now maybe the AI could be programmed to take all of this into consideration. That could take a lot more programming and a lot more of a hassle for our programmers.
And the AI takes a lot of CPU power, so if we have too much AI it would likely be bad. Plus like Drek says: "there's always a better alternative to AI control systems and to sliders" (something like that, lol).

How many games have actually had AI control systems that worked 100% to the players liking. I bet there are few, if any that work very well (Moo3 is not at the top of the list of good AI Control systems).
Hopefully you realize that I was trying to say that I didn't like his idea. I am still of the opinoin that Phase-time is the best.
200 and still a Wyrm!?! I don't want to be a Wyrm anymore. I've been a Wyrm for 100 posts now.

User avatar
Daveybaby
Small Juggernaut
Posts: 724
Joined: Mon Sep 01, 2003 11:07 am
Location: Hastings, UK

#162 Post by Daveybaby » Sat Oct 16, 2004 11:49 am

MisterMerf wrote:In some cases, those two things will be inextricably linked. For instance, let's say I can give my AI an excellent state to work from if only I run an algorithm that bears a little too much resemblance to the knapsack problem. Now all of a sudden, I'm limited to 15 task forces per battle before it starts taking years to update the AI's conception of the battle.

In choosing AI strategies, there will also be a lot of algorithm design that would surprise me if some of it didn't have to choose between "intelligence" and performance.
I'm just saying that the limiting factor is not processing power, its the amount of time you can spend developing the algorithms. Its not like we're training neural networks here.

The only exception to this is in things like chess AIs, which basically play the whole game through millions of times for each possible move. Not an approach which is viable for a moo-a-like.
The COW Project : You have a spy in your midst.

User avatar
haravikk
Space Kraken
Posts: 104
Joined: Wed Mar 31, 2004 10:04 pm

#163 Post by haravikk » Sat Oct 16, 2004 12:34 pm

@Ranos

You misunderstand, the idea is more of an addition to real-time combat to allow it to be simplified, this will be especially useful for large, extremely active or spread out battles where the player can't easily move around to issue each order required of them.

It gives them more of a sort of 'comand post' or 'flagship' command position where you really just tell your TF's what to do and they'll try and do it, seeking advice if they run into trouble.
But if you really want or have to then you can still issue orders to TF's that think they're doing just fine, it just means moving all over the place to do it, where a TF that is requesting orders would add to your list, and when you click it will pop-up a box with a brief overview of it's status and mb a mini-map showing what is nearby so you can quickly make a decision without moving your view from where your 'camera' is at the time.

This would also make the 'bullet time' idea I suggested along with it more intuitive. The bullet time would kick in as more situations arise and are added to each player's list, rather than being based on frequency of weapon's fire or such.

User avatar
noelte
Juggernaut
Posts: 872
Joined: Fri Dec 26, 2003 12:42 pm
Location: Germany, Berlin

#164 Post by noelte » Sat Oct 16, 2004 8:39 pm

The only exception to this is in things like chess AIs, which basically play the whole game through millions of times for each possible move.
That's proven impossible. :-P But i got what you meant.
Press any key to continue or any other key to cancel.
Can COWs fly?

Ranos
Dyson Forest
Posts: 234
Joined: Tue Mar 23, 2004 6:24 am
Location: Northern Wisconsin

#165 Post by Ranos » Sun Oct 17, 2004 12:37 am

haravikk wrote:@Ranos

You misunderstand, the idea is more of an addition to real-time combat to allow it to be simplified, this will be especially useful for large, extremely active or spread out battles where the player can't easily move around to issue each order required of them.

It gives them more of a sort of 'comand post' or 'flagship' command position where you really just tell your TF's what to do and they'll try and do it, seeking advice if they run into trouble.
But if you really want or have to then you can still issue orders to TF's that think they're doing just fine, it just means moving all over the place to do it, where a TF that is requesting orders would add to your list, and when you click it will pop-up a box with a brief overview of it's status and mb a mini-map showing what is nearby so you can quickly make a decision without moving your view from where your 'camera' is at the time.

This would also make the 'bullet time' idea I suggested along with it more intuitive. The bullet time would kick in as more situations arise and are added to each player's list, rather than being based on frequency of weapon's fire or such.
That makes sense. It would allow for the normal control of real-time withour the hassle of trying to keep you attention fully focused on all locations at once.

If real-time is implemented, I would like to see this used, but I am still in favor of Phased-Time.

I'm not sure I like the idea of the bullet time though. This would cause each battle to last longer. Either that or it would mean you would have less time for each battle, depending on how the bullet time would work.
200 and still a Wyrm!?! I don't want to be a Wyrm anymore. I've been a Wyrm for 100 posts now.

Post Reply