Guidelines for smart units

Discussion in 'Planetary Annihilation General Discussion' started by dmii, October 4, 2012.

  1. dmii

    dmii Member

    Messages:
    138
    Likes Received:
    1
    I have seen multiple suggestions containing the call for "smart units" or "less dumb units".
    While I agree with having less micro-intensive gameplay, since that would be impossible with the scale PA aims for, some suggestions are going way too far for my taste.

    Since I am talking about guidelines to judge, whether something goes too far or not, my taste admittedly is not relevant. However, while I was thinking about why exactly I don't like some suggestions, I came up with some points and reasons, which I think a lot of people can agree on.


    Well then, here they come (Examples follow after them):

    1. Units should only automatically do a task, if it always benefits the player, no matter what the situation looks like.
    Reason:
    Nobody likes to lose because his units decided to do something bad. Not having to constantly keep track of your units is what having smart units is all about. You want to be able to focus on what the game actually is about and not have something else constantly interrupting you. If you can't be sure, that your units are not screwing you over, you will have to check what's happening to them more often than necessary. In other words: They don't achieve what smart units try to achieve.

    2. How smart the unit is must not affect balance in a significant way.
    Reason:
    Balancing is always a difficult task. However, if the behaviour of the units becomes a balance issue, you open a big can of worms. Normally you can balance units by adjusting values, but how can you balance behaviour? Tweaking how good units are at automatically doing something isn't a simple change in a value, it may require a completely new algorithm.
    Additionally, unit behaviour can distort the importance of the other values you have to balance, making it more difficult to judge the effect of changing them.
    (I think this point becomes more clear with examples, so bear with me for a little.)

    Examples:

    A good example for the first point is automatic retreating. Imagine two armies fight and an artillery strike hits one, knocking a lot of units down to the point where they start to retreat. The result is, that a part of one army starts retreating, which causes them to be outnumbered, most likely leading to them losing the battle, or also retreating shortly after, since the damage the opposing army deals is focussed on less units.
    Any situation, where a group of units gets evenly damaged over time, can lead to this scenario. It can even make an unfavorable battle turn around completely through the shift in numbers.

    For the second point automatic kiting is a good example. Having less speed and range can in extreme cases make some units completely worthless since they never get to attack. Depending on how good the units can kite a speed or range nerf/buff can have a very big effect, especially if units can move and shoot at the same time.
    Automatic dodging is an other example for something you can't simply change and is also heavily affected by the unit-speed, since sluggish units usually don't have a chance to dodge anything and fast units could be too speedy to hit. (Admittedly, you could change the dodge algorithm to randomly not dodge, but that runs into the problem of you having to check more often, to see if everything is fine. Which defeats the point of having smart units.)


    Before the closing words, there is something to mention: Some of the examples were -as far as I remember- suggestions, which allowed the player to toggle the behaviour. As I already mentioned, the point of having smart units is, that you can rely on them always doing something beneficial for you without checking on them all the time.
    While toggled behaviour doesn't conflict with the guidelines I propose, since it makes the unit behaviour a player decision, it still has the problem of the player having to regularly check wether the behaviour is still optimal. Which again defeats the point of having smarter units, so toggled behaviour shouldn't be used too much for the general unit AI.
    Though I could imagine having something like kiting as a toggled passive unit ability for some units. (That could be an idea worthy of its own thread though.)

    So there we go, I hope this leads to more suggestions, which actually lead to people having more time to focus on the grand scheme and don't have a possibility to make players have to fight against the AI of their units. Because in my opinion, we don't have a lot of them and more of those which substitute the problem of units not doing enough with units doing too much.
  2. thorneel

    thorneel Member

    Messages:
    367
    Likes Received:
    1
    If I understand correctly, you means that the 'smart' units shouldn't do something really dumb because of a hole in how their algorithm works. So yes, obviously, the unit's algorithms should be tested and problems like that should be fixed.
    If during the test, it appears that units retreating causes your army to have twice more losses, it means that the retreat algorithm is flawed, and that a new retreat algorithm should, for example, make the whole army retreat.


    No, it's the opposite actually. Balance should be made by taking into account that the units are smart. If one unit is worthless against another because it is slower and has less range, it means that the second unit is a great counter to the first one, and the game should be balanced accordingly.
    For example, maybe the second unit is a skirmisher, shooting slow projectiles from long range. Maybe the first one is one specialized in assaults against structures, slow with lots of HP, or maybe its specialized against swarms, with fast-firing and/or AoE weapons. It means that against one of those, you are meant to send the skirmishers. And it means that you have cover those assault units with, for example, fast raiders that can take the skirmishers out.
    So first make the units smart, then balance the game accordingly.
  3. thetrophysystem

    thetrophysystem Post Master General

    Messages:
    7,050
    Likes Received:
    2,874
    About the guy above me's post, it makes sense to send large-scale war machines with fast scout units alongside. If you run into kiting units, you would attack them with faster units and try to pull back the heavy units and surround them or circle-jerk them to get them INTO range of the larger AOE units.

    So, less one-unit-spam warfare, more diverse armies and different units doing different jobs and flanking different ways.
  4. dmii

    dmii Member

    Messages:
    138
    Likes Received:
    1
    I am more talking about not implementing a behaviour if it can lead to units making bad decisions, than trying to fix bad behaviour. Because even with testing you can never be sure, that the behaviour actually works as intended.

    My point isn't that the AI shouldn't be taken into account, it is, that it shouldn't make balancing even more difficult than it already is.
  5. sylvesterink

    sylvesterink Active Member

    Messages:
    907
    Likes Received:
    41
    First off, you disregard something like unit retreat far too easily. You are correct in the fact that in a major battle, where you have a prepared army, it would be foolish to have them automatically retreat. A few unit losses are worth an overall victory in this case. But if you have a small group of scouts/raiders that are out harassing the enemy's economy, or doing a minor strike, then if a counterattack comes that would wipe out the group, you do not want them to stay there and get wiped out. Normally you would personally order them to get the heck out of there.

    But suppose you're going to pop over to your base and set up some defenses, or maybe go to another planet and organize an attack there? You won't be able to hold the hands of your units that are raiding, so what should you do? Cancel the raid and avoid losing units? Accept that the entire group could be wiped out with no tactical gain? Or what about a third option: toggle the units to retreat when they reach 60% hp and send them on their way. If the raid works, then you've achieved a tactical gain. If it fails and your units get attacked, they run and you incur fewer losses than if they didn't retreat. And supposing you did come back to monitor those units as soon as you heard of the issue via a report, you can easily select the units, disable retreat, and give them other orders that may give them better survivability (say the retreat path is blocked), or, if their situation is hopeless, allow them to do as much collateral damage on their way out.

    This does not reduce player involvement in any way. It is instead a tool to allow the player to have MORE involvement and more options for dealing with situations.

    Then there's the idea of unit micro, primarily in the form of unit kiting. If the units are unbalanced in a way that kiting would make one set of units dominate another set, then what would prevent players from doing it manually? Now, suddenly the advantage goes to those players with the 1337 APMs. The imbalance not only remains, but suddenly it's not just an ingame issue, but a metagame issue.

    The same goes for swarming. I agree that units shouldn't perfectly dodge projectiles as they are attacking, but having the units attack in a swarm of movement results in slower projectiles becoming ineffective on fast, agile units. Suddenly a few Peewees are not useless against a lone Bulldog, so providing protection for that Bulldog becomes a necessity. Weapons that fire fast or have a solid aoe will more than make up for it. And even still, that lone Bulldog will get some hits in, because the units are swarming, not dodging, which makes sense from a TA perspective.

    Of course, sometimes you don't want to kite or swarm, in which case turning it off is an option. (OMG, not the dreaded toggle!!! ;-) ) Again, much like toggling retreat, this isn't a detriment as it offers the player more options.

    And once again, I have to point to Zero K. The game implements all of this stuff right now, so if you want to see whether these features break the game, play it. It's like having a mockup of those features available, so instead of arguing against a principle, argue against something you've actually tried.
  6. theavatarofwar

    theavatarofwar New Member

    Messages:
    84
    Likes Received:
    0
    Lets set up a typical use case, and see how we think it should go.

    Lets say both sides enter a given theatre of war with a small group of units with the purpose of engaging each other. No large objectives like "push past to take out the shield generator of those rebel scum", just units versus units. And heck, lets further assume that both groups have an identical mix and quantity of units.

    Who should win, and what should determine the outcome?

    If neither player is paying any attention, it should be a draw. The units will engage, do whatever units are preprogrammed to do, and likely one side will win (could be either side) with just a couple units left. In this scenario, smarter AI wouldn't do a thing. It would make no difference if artillery can keep range, or if light tanks can serve as point defense and interference, because both sides would be operating similarly. In this situation, anything that is done by AI is moot.

    So what if both players are paying attention? In this situation, advanced AI isn't needed either. Each player can control the units and give them commands for what they feel is best for the situation, and the winner may be determined by which is the better strategist, or more capable of splitting attention. Is it micromanagement to sometimes select individual tanks and issue move+attack orders? Absolutely.

    What if one player is watching the action, and the other one isn't? Now we're down to opinion. To be perfectly clear on this, my opinion is simple: the human should win, period. He shouldn't need an optimal strategy. He shouldn't need to decide what role each of his units perform. He shouldn't have to manually dance his units around for optimal placing. Because the goal isn't to beat the computer, the goal is to beat the enemy human player. If the units have advanced AI, and the opponent can handle every unit on the unit level with very specific decisions (this one back up to optimal range; that one advance to soak enemy fire), now the human has to be quite skilled in understanding the AI and how to beat it. Meanwhile, the enemy who isn't even looking glances over and sees "oh look, my units fought something and won", and moves them on to another objective. This is not a competition anymore; both players may as well be playing solo games against the computer.

    In order for the human player to matter, he has to be able to make good decisions as well as bad ones, on every level. The more control the AI gets, the less decisions the human needs to make.

    Where do you draw the line? As low as possible. Remember attack+move? Technically that is AI, so should it be removed? Personally, I'd say no. attack+move solved an existing problem, where two enemy groups could wander right past each other without either side knowing they were ever there. Thats no way to conduct a war, especially if one of the groups is the starting humvee that GDI always gets, loaded with an engineer that can be built 60 seconds into the game, and is marching straight to your construction yard (C&C players here?). Well, should we make the AI good enough to not only shoot back, but to abandon their current orders and pursue the unit thats trying to drive by them? Nope, now we're taking strategic choice out of the hands of the player, and into the third scenario listed above. Well, should the humvee be able to zig and zag on its own, to avoid the pursuers (assuming they pursue because a human told them to)? Nope, because now its the AI versus the player again.

    The current system is pretty good as it is. Pathfinding tweaks? Sure. Decision-making tweaks? No please.
  7. dffmmm

    dffmmm Member

    Messages:
    56
    Likes Received:
    0
    I thought this game was about robots with only one directive: to krush kill 'n destroy everything thats not them. I don't see how retreat fits in there anywhere. They just keep on going until all else (or they themselves) is no more.
  8. dmii

    dmii Member

    Messages:
    138
    Likes Received:
    1
    I am actually not against toggled stuff, I just don't think it should be used a lot ;)
    That paragraph is mainly there to prevent people from just making everything advanced toggled and not at least think about it.
    The more I think about it though, turning advanced behaviour into a unit ability may be an other thread to create, which could eventually turn out to be a gold mine :D
    Mainly because the advanced stuff is quite useful, but also situational, and some units may not even benefit from it at all, so it is completely useless to have the behaviour on them in the first place.
  9. exterminans

    exterminans Post Master General

    Messages:
    1,881
    Likes Received:
    986
    Nice example, yet you forgot one very important parameter: The size of the battlefield.

    If the battlefield is to small, then proper reactions require also fast reactions, most likely to fast for any decent human player. If you have a LARGE battlefield, then advanced tactical decisions like flanking, ambushes and alike come into play and might / will lead to a different outcome.

    While i share your opinion that the human, active player should always be in advance in the latter case, the first case should never(!) be in favor of high APM players!


    The features most asked for, like kiting, swarming and intelligent target selection, won't collide with any commands on the level you would normally give in a game of this scale. They do help you though to cap the benefits of high APM at somewhere about 20-30 APM and shifting the focus of the battle from intense micro management to high level tactical decisions.

    Latter ones are highly underrepresented in most modern "RTS" style games, one of the few exclusions being the medieval scenarios from the Total War series.
    Problem in most games is the fact, that micromanagement can give you an advance of 100-200% over the stock behavior while larger scale tactics are usually pretty much useless since most games are to fast paced to make use of them - a result of ridiculous small battlefields compared to the movement speed and DPS of units.
  10. dmii

    dmii Member

    Messages:
    138
    Likes Received:
    1
    Just to debunk this misconception: Micro doesn't favor players with higher APM. It favors those who control better. A high APM is the result of players controlling a lot, however, whether they do it good or bad has nothing to do with how much they click.
  11. exterminans

    exterminans Post Master General

    Messages:
    1,881
    Likes Received:
    986
    When dodging missiles with an insane amount of movement orders, it IS APM what gives you the advantage. (The advantage is close to infinite in this case)

    Same when swapping the order of units in the formation to prolong the durability, that's also something where only APM matters. (Effective durability of the squad and therefor total damage output can be enhanced by 50-100% this way.)

    Also when moving units always slightly out of enemy range when two unit's ranges differ only slightly and the unit with the higher range is also faster. (Again, the advantage is close to infinite and high APM are required.)

    Or when calling back units which started auto-chasing to prevent them from running into an obvious trap. (Your enemy abused the stupid AI behavior and you actually require high APM to achieve the desired results now because you had to turn off auto-chase completely.)


    Just to give a few examples and these aren't the only ones.
  12. dmii

    dmii Member

    Messages:
    138
    Likes Received:
    1
    First: All these depend on how exactly the engine, units, AI and the UI work.
    Second: APM is a number, which tells how many actions you do. It says nothing about wether the actions actually were useful. A high APM means nothing if you can't put it to use. Your useful actions are what give you the advantage. Those are where the skill is at, high APM in itself is just something people obsess about, which actually doesn't say anything at all.

    On a side note, in a lot of games, a simple U-turn is enough to dodge a projectile. That's one click and makes this a perfect example of how APM doesn't say anything at all.
  13. jseah

    jseah Member

    Messages:
    129
    Likes Received:
    2
    There was this vid on youtube with one of the BA uber-micro cases where a single AK dodges the fire from about ten rockos for a full minute (and kills two of them).

    Since rockos will lead their target, the only way to do that is to change the move order every time a rocko fires. If that isn't high APM, I don't know what is.


    Also, I suppose we disagree here, I don't consider *execution* to be a skill that an RTS should test. Decisions, yes, executing them, no.
    I do not want to fight the interface, an ideal interface would understand exactly what I want my units to do and make the units do it. ("form concave from left and deform where approached, do not wander beyond this tolerance... etc.")
    The easier it is to perform the various actions, the better the interface and the better the game.
  14. dmii

    dmii Member

    Messages:
    138
    Likes Received:
    1
    It is high APM, but that's a result of the microing guy actually doing a lot of work and not simply spamming commands. High APM is something which often comes with good micro, but it is no indicator of it.

    How much it should be rewarded is not something which really bothers me, I agree with not having to fight the UI, but I also don't want units playing for themselves. Which is what I have seen suggested too many times and the reason I opened this thread. (And I hope it gets more back to the original topic now xD)
  15. lirpakkaa

    lirpakkaa New Member

    Messages:
    60
    Likes Received:
    0
    More like 20 seconds, and kills one that's already on low health.
    http://www.youtube.com/watch?v=ROpuB9KBsBk
    That part is around 4 min mark
  16. jseah

    jseah Member

    Messages:
    129
    Likes Received:
    2
    Oh, wow. That definitely felt longer when I was watching it. >.>

    Fine, I stand corrected.

    Although I probably should mention that I can do roughly half of that (dodge 5 rockos for 20 seconds without killing any) at the cost of not doing anything else in the game. The moment I look away, the AK dies even with auto-swarm.
  17. lirpakkaa

    lirpakkaa New Member

    Messages:
    60
    Likes Received:
    0
    But if you want to eliminate the need to exactly perform what you want, wouldn't the most sensible way to do it be to just make the game simple? Making a simplistic RTS with some transparent RPS mechanics and good scouting opportunities, would be the ideal then - all you have to worry about is the high level decisions and mind games.

    I don't see a point in having complex physics and economy and all that jazz, if the AI will take care of all the low level management - that just obfuscates the high level decisionmaking, while the low level only offers a fireworks show to watch and not much more, while taking tremendous amount of dev time to balance and then build that AI that plays the low level game.

    Now if the AI/UI only is so good that you still have to give it complex orders like you outlined... It doesn't really remove the execution part at all. You get more precise tools to do what you want, that means that to stay relevant to the competition you have to be more precise as a result. You just have to decide, which you like more - execution of picking up the right AI modes, retreat areas, no-go zones, etc., before a battle, or just straight up handling units during battle.
  18. jseah

    jseah Member

    Messages:
    129
    Likes Received:
    2
    A simple game simulation is a game that doesn't have depth. Just take a look at what happened with BA's not letting you shoot through allied units colvols (shots collide with friendly units)

    Suddenly you had battlelines. And then you had flanking tactics like "Crossing the T". Source; see Maneuvers near the bottom.
    The tactics are *interesting* decisions. The line itself is NOT an interesting decision (its the default behaviour of anyone who played any BA in multi), the evolutions of tactics from the line formation is.
    The UI power of the custom formations widget that gave us line move commands allowed this to happen; can you imagine playing TA when units would block each other's shots? or starcraft? >.> now that would be hellish.

    You don't take the mechanics away, do not reduce complexity unless the hardware cannot handle it. You improve unit behaviour and UI to adapt to the mechanics. The decisions that fell out of BA's stricter friendly-fire simulation could only be implemented successfully with the right UI.
    As hardware improves, you can afford better unit behaviour algorithms, more powerful UI and deeper, more complex, simulation mechanics.
    Which all lead to additional interesting decisions and interesting decisions with more depth (harder decisions).


    WRT to autokite:
    Auto-kiting skirmishers/artillery allow them to scale to multiple engagements by making their player-attention requirement more similar to the plain assault tanks. This means that the natural counters of slow and low range units (but high hp/dps) that are these skirmishers will hold to a larger scale, instead of reversing when you get to the level of throwing armies at each other along six different paths at the same time.

    This preserves the complexity of a tactical single combat into the larger scale; where if you had too many things to do and couldn't focus down on every group and then units just sat around shooting until all enemies in sight were dead, it would just favour spamming a mix of high hp tank and a high DPS bruiser for field engagements or they will end up being worth less than their cost or take your attention from somewhere else... like vying for air superiority.

    And both of you ought to be able to have both happen (start a skirm duel and fight an air battle) simply by making the decision for it to happen. "I want the units to kite like *this*". Obviously, we're never going to reach the ideal UI that does this, but we can improve towards it.

    The tool (auto-kite) allows the player to use mixed armies to greater effect, adding additional interesting decisions. (because now you can add skirmishers to large maps when they were less effective and have to respond to his)

    The kiting itself might be interesting at a low level (skirm vs skirm duels is like playing whack-a-mole), and YES, auto-kite takes that away. But the consequences of a small skirm-skirm duel quickly approach irrelevance when you get to macro scale; it isn't as interesting anymore. You just want them to skirm, and it does not even have to do it like a human, just some would be alot better than none at all.
  19. lirpakkaa

    lirpakkaa New Member

    Messages:
    60
    Likes Received:
    0
    Yeah I agree with a lot of that. However, it's very much in contrast to what you said earlier:

    Because changing the control scheme into a more complex and powerful one, as long as units are not autonomous, does not remove the need to have precise input into the game by any means.
  20. jseah

    jseah Member

    Messages:
    129
    Likes Received:
    2
    ...
    Do you get the feeling we are arguing over definitions? What do you think of when I say "execution"? And what do you think is a "decision"?

    I consider a "decision" to be some committment to a goal. (There's a different set of things that I also consider a "decision", these relate to guessing at unknown information, the value of information (usually also unknown) and future states; so not relevant to this)
    Shooting a rocket is not a decision. Its execution of a decision "I should blow up this target". The player does not tell his units to shoot rockets. He says "kill X".
    Ok, that was trivial. Pointless even.

    What makes decisions "interesting" is that they not be trivial. Trivial decisions are those that are not solvable satisfactorily and have significant impact.
    MM and energy balance is roughly solvable. You won't get optimal, but you can get very close with a very easy algorithm (E > X, turn on; E < X, turn off). Clicking the MM and (de)activating is execution; worse, it is the execution of a trivial decision (E too high! turn on MMs!) because the problem is easily to solve satisfactorily.
    One of my rockos is out of line by one rocko width. Solving it is more difficult (for AI), especially if all of them are slightly out of line (and the line hugs a cliff say). Now that hurts efficiency of the line... a bit. Using a move order to send it back is the execution of my decision to have a line (which is a trivial one, since lines are best for rockos in nearly all cases), but I won't bother since it has little impact.

    Deciding to use a Shield and Rapier style attack (mentioned in the link in previous post) *is* a non-trivial decision. When to use it is not easy (at least if you haven't been fighting on the BA/ZK frontlines for years), and using it nets significant gains (or significant losses if you used it wrongly).
    Execution of this is giving a line command to assaults, followed by a line command to the raiders a bit later. If you could somehow simplify it to a single command (eg. select both assaults and raiders and click "Shield and Rapier" followed by a line), then why not?


    EDIT: I hope that makes it clear what I want RTS to test (decisions) and what I do not want it to test (execution). Knowing what commands to give should always trump simply doing them in importance. And trump them by a whole lot.
    EDIT2: this means that Best decisions executed well, should win over slightly less good decisions executely perfectly. Obviously, when both of you make the best decision possible in an equal state (unlikely), then the person who performs the best execution will win. But that sort of situation should be made unlikely, the game should be biased towards creating localized differences and make the *best* decision very hard to know so you never have the same set as your enemy and never know if what you just decided to do was really the best.


    ...
    Hm.

    Actually... why not? Presumably the difficult to solve variable of "how do you time when to send the raiders" might be acceptably approximated by some delay based on the distance of the order...
    Hrm, I shall have to think about that. Maybe a new suggestion in that thread will go up soon... =D

Share This Page