Better Commands; specific ideas

As a matter of fact: No, a halfway decent player is not sufficient to beat an state of the art AI. Maybe when it comes to strategical decisions, but the AI is superior in base construction and when it comes to tactical decisions like target selection, kiting and intercepting.

This is why the use of AI assistance must and will be limited to mods which have to be accepted by the server, pure client side UI mods won't get access to the required controls.

I feel the need to mention that ZK and BA both allow widgets that can effectively play the game for you.

No one has felt the need to replace themselves with a bot.

Additional suggestion from reading the thread on power drawing weapons:
Resource Reserve Levels
ZK recently implemented this. You can click on your resource bar and that amount of resources will be reserved for high priority construction and basic running of units.

Essentially, the reserve level is a crude way of saying "power that radar and shield before using it to build". If you don't have enough E, your radar and shield stays online, you just build slower (and start to pile up M)

It could be interesting to have something like this, or even a system that lets the player set what receives priority power, allowing the diverting of power to priority projects? just brainstorming out loud.

you don't need a "face" command. you just give 2 movement orders, in the shape of an "L" lol.

Sometimes you want them to turn in place though. Especially when you are giving line commands when you cannot be sure that you can replicate the exact same order but displaced slightly.

I see this being especially useful for naval combat.

But mostly the thing it does is make microing more simple (and not in a good way). When there's some "stupidity" in the targeting,

For example those units with big damage, long reloadtime guns, if they never shoot the same target etc., you have to balance to account for that - they cannot be as powerful as with a usual targeting system. But if they are really powerful but slightly dumb... It opens up microing possibilities for both sides, to trick several of them to shoot a cheap scout, to position them so that they won't waste shots on 1-shotting the same opponent. This is all cool stuff, yet not frustrating stuff by any means. When the battle becomes big and hectic enough, you just let them handle themselves then, and they still will do fine. Or maybe you'll decide to take units that perform better in a big battle where you can't babysit single units.
This kind of targeting otoh, makes all units perform more samey. And following that, choosing the right unit for the right situation is less crucial and interesting. Units acting differently when in different numbers is fun, both strategically and tactically!

The sending a scout to distract AA fire is a good example. You might think it's stupid, but what do you gain by removing this tactic? It's easy enough to pull off mechanically, anyone can do it. And it makes the air game more diverse. You plan your turret placement to account for this, if you cannot draw fire like that it makes turret placement less important.

It's behaviour optimisation just for the sake of itself, it does not lessen the mechanical learning curve, or add tactics to the game. Rather it removes some tactics in the name of sterile "better behaviour".

But why should the tactics of a game hinge on deliberately dumb behaviour of units? The argument against it is analogous to the argument of why we shouldn't have a UI that limits our actions.
EDIT: and then you end up with the high alpha, long reload guns, balanced for being less useful in larger engagements, turn into an unstoppable OP unit in the hands of a 400 APM player. Who is essentially doing the exact same thing this targeting control will do, only more limited.

In case my point got lost in the wall of text (I admit I was getting a bit... excited there), I want the weapon types to scale better. This is one way to preserve the difference in weapon types from small scales to big scales.

A high alpha weapon with some shot flight time wins out in small engagements and loses in large engagements, compared with DoT weapons (which are more flat). This cooperative firing control will tend to make weapon characteristics more scale-independent.

A high alpha weapon is still a high alpha weapon, and will perform to its expected characteristics (lots of initial kills when entering range) compared to a continuous DoT weapon if this system is implemented, regardless of the scale.

Did I miss some major development in game AIs?
Maybe some researchprojects are working on better AI's but there are no AIs for any RTS I know about that can compete with human players. Usually the only AI's that can compete use cheats. Also an AI is incapable of dealing with the developing playstyle of human players, it cannot learn new strategies.

there is one, and only one i am aware of. some uni in the states made an AI for the original starcraft that beat spain's #1 player apparently. but of course this was more than a decade after the original game was released.

agreed, RTS AI's suck. and that's no criticism of developers. AIs are just really hard to make.

Well, why not? As long as it is diverse and tactical and not sterile and samey, is it a problem what paradigm it depends upon?

Not exactly, no. It's a different thing when you make something that you'd be doing anyway, easier, than to go and optimize behaviour the player was never gonna touch directly.

I don't see it as being so friendly to people with crazy good mouse accuracy, as to people who are just good at thinking forward and positioning. Sure, maybe someone will make their units to hold fire and personally target each shot for maximum precision, while simultaneously keeping track of their army movements, scouting, and economy... But for non-superhuman people it's more about prepositioning in a way that you spread your low RoF units that they won't shoot at the same moment, sending in the scout before your army to bear some of that frontloaded damage,
Things that require tactical thinking and planning, not dexterity.

If you do not like small-scale tactics and want a totally economy-oriented game, you might as well make the units simple enough that the targeting algorithm didn't really matter. But if you want some fun tactics, avoiding overkill by AI is not optimal.

Quoting APM counts is a bit problematic. Someone who has good APM is likely very in tune with the game they're playing, keeping check of stuff in every corner of the map and not forgetting their ****. This is the epitome of a good RTS player.
But then often when people refer to them crazy APM people, they think of manual dexterity more than anything.

But why do you want that, exactly? I don't see the problem when some units are better in small skirmishes, others in large battles. It's a good thing more than anything. Rewards the player who can force the kind of engagements their units are suited to. If everything scales well between battle sizes, that's boring.




Here's actually a piece on this exact subject that I saw not long ago.
http://www.teamliquid.net/blogs/viewblo ... _id=371077

Skip to the "So what is so interesting about Overkill?" part for maximum relevance.

It wasn't Spains current #1 player, it was some dude who had played the game at WCG a decade ago and turned mostly inactive since then.

oh OK, thanks for the correction. still pretty impressive from berkely uni though, but also reinforces the point that RTS AIs cannot compete.

Wrong, they can compete when focusing on tactical decisions on a micro level. And thats the type of decisions we are talking about. When choosing targets, joining and parting in attacks, the perfect decision CAN be calculated.

Only strategical decisions are still an issue as normal AIs are incapable of actually GUESSING which strategy the enemy will go for. Another issue is flanking and alike advanced tactics above the micro level, as such tactics require far more information about about the situation then the AI could handle.

Thats why Starcraft AIs are capable of beating human players, because the game focuses solely on micro while strategical decisions play a minor role or are close to non-existent.

give us an example. i am yet to experience such an AI, and i've been playing games for more than 30 years.

PKC:
The Berkeley Overmind was an AI that used mutalisks in a spread out fashion to avoid AoE which was the usual counter. If there was a UI for SC1 like spring has, a human would probably easily crush it (provided the game was rebalanced for that, coz SC was balanced for a weak GUI... T_T)

Additionally, starcraft is a kind of game where orders and commands have immediate impact. Maneuvering, movement and terrain are all black and white. In TA and Spring (and to a lesser extent, SupCom), the simulation is more detailed and messy, feedback on what you did wrong was a bit harder to see and relies alot on pattern recognition, which AI totally fail at.

While building the AA firecontrol gadget, I found myself running into limits on computers (can't order a ray trace for every target in range for every tower) and limits in the amount of logic I was willing to make for it.
Client side widgets, if they contain significant AI, will be built to work *with* humans, not to take over a human's job.


lirpakkaa:
Now come on, the best ZK players are able to keep tabs on every area of a large battlefront at the same time, I'm willing to bet that in the "high elo" battles I mentioned in the intense frontline thread, the enemy couldn't have attacked at any point along our line without all of us being aware of it and what our allies' response was and if we needed to pitch in.
None of us ever get more than a hundred APM? At least I don't think so. (I personally don't reach much more than 60, perhaps 80 on an ideal game. Although I am not a top rank player, just experienced. Maybe equivalent to a Silver in SC2?)
We keep track of the strategic situation easily. ZK's GUI is very friendly for that sort of thing (strat. zoom, intel overlays, pathing/heightmaps, weapon range circles, show all commands including allies, unique but clear unit icons, etc.)

I still manage to miss things (probably why I'm not a top rank player =S), but clearly high APM is not a result of being in-tune with the game. High APM is the result of constant practice at individual tasks and incredible mouse accuracy.

I read the article. I don't see how the decision to spread tanks is at all relevant. Spreading tanks in a Spring game is a single command (longer line command that bends back), spreading *buildings* is a matter of dragging a line, tapping z a few times (increase spread) and pressing alt (for box area building, or was it ctrl?). It is only a relevant decision in SC because the micro requirement for doing it sucks your ability to pay attention elsewhere. If you could spread your tanks and assaults (and send decoy targets) with just a few clicks in SC, that would just become the default way to use them.
Sure, you wouldn't clump your tanks if the same problem was presented to you. But if you accept that we should have powerful UI, then it isn't an interesting decision, just one more thing you have to memorize and remember to do.

Baiting with fighters has been mentioned as too easy to execute and it makes Screamers in ZK nearly useless (people will just put a gunship factory on repeat suicide drones into your range to drain its ammo).

As for differentiating weapon types, high alpha weapons remain high alpha weapons, even with this fire control. If you have a bomber problem, you still want the high alpha low DPS missile towers. If you have a gunship swarm to deal with, you still want the high DPS continuous DoT flak guns (AoE is bonus).
Same with ships. Enemy spamming cheap frigates? You want lots of AoE and rapid fire cannons. Enemy building capital ships? You want high alpha missiles or torpedoes.

I doubt having a AI that could beat top level rts players is impossible, it's just that there isn't much academic research about the AI of any specific rts game yet.
On the macro level, I don't think the most rts games have more strategic complexity than the chess or the checkers.
On the micro level, sure the rts playgrounds have so many more legal positions than the boards, but the AIs could have superb fast reactions and almost unlimited APM, it's a big advantage over the humans.

I wouldn't be too sure about that.

Chess has perfect information. RTS games do not. The value of unknown information is a whole different area of strategy.

Yes, more legal positions than in chess, but at the same time less complex because you can resolve tricky situations much easier than in chess as you can move all units at the same time.

A modern RTS AI works similar to the classic database based chess AI, it has a lot of experience and will try to match the current situation with one it encountered earlier. And though the number of possible games is close to infinite, neither is the number of possible moves per situation nor is the number of realistic situations.

It will always choose the most important unit as target (because thats what the AI was rewarded for in training), whereby the importance of a certain unit depends on the environmental parameters, like much fire support does it have, is it capable of protecting other units, can it deal much damage (sniperbots and rocket launchers, but also artillery, nuke and KEW), do i win the game if i kill it (commander).
Any modern AI is capable of making the perfect decision for the current situation based on a filtered set of statistical data which is then run trough a neural network which has been automatically trained with thousands of combat situations. The AI will show the behavior it was trained (rewarded) for. That can be "achieving at least equal resource loss", "surviving with minimal losses", "wipe the enemy out" (aka attack if you have chances to win or at least to achieve a draw, no matter the losses).


You want an example? Try the tactical Sorian AI in Supreme Commander 2 with the latest patches, playing on the highest, non-cheating difficulty and try to counter an attack from the AI with an equal number of units of the same type.
Don't use advanced macro tactics like flanking or abusing glitches in path finding and you WILL loose. No matter what. It just outperforms you by far.

You can still just call in reinforcements of the perfect counter type to wipe the AI of the map (it will start to flee if it sees any chance of rescuing at least SOME of it's units!), but thats not the point. Or more important: Thats not what is supposed to be automated as almost every player should be able to handle such situation on his own as it doesn't require dozens of APM, also sacrificing a group of units instead of risking another one can be considered a strategical decision.

There are some very powerful AIs for the competitive Rock-paper-scissors, which is a complex game about guessing the opponent's behaviours, unknown information plays a much bigger role in that game than any rts.