|
Post by boosters on Sept 29, 2016 17:02:54 GMT
Hi everyone, I think anyone who's been on these forums the past few days is probably interested in figuring out how to win space battles, and I can think of no better way to do that than to test those designs against each other. So, I've been thinking of a way for us to do that, and I think the following might work: we design a fleet, describe its composition and its desired AI behavior here, and upload our UserDesigns.txt so that other users can try it out. I'm thinking that we need some kind of "weight class" thing going on, and some initial battle conditions, so I am initially going to suggest the following rules, just because it sounds fun to me: "MARTIAN THROWDOWN" RULES---------------------------- 1. A 20kt mass limit for each fleet. 2. A 500Mc cost limit for each fleet. 3. Battle takes place in Martian orbit. 4. You control your own fleet, and the other user's fleet is under AI control. Each fleet posted must include the AI behavior you desire it to have. 5. Best of three combats. ---------------------------- You test your own fleet in three rounds against each of the other users' fleets. The user whose fleet wins the most battles across gets gold, with silver and bronze to the next two respectively. ---------------------------- I am, of course, open to suggestions and updating these rules if something else sounds more fun.
I think that one possible problem with these rules is that the AI doesn't know how to command your fleet optimally, but hopefully that will be mitigated by the fact that you are also flying your own fleet on your own machine against all the other competitors' fleets.
If the AI sucks at winning for you, you can just shake your fist at your incompetent space captains.
|
|
|
Post by quarkster on Sept 29, 2016 17:20:42 GMT
Where will the combat take place? What kinds of initial orbits? These are important design factors.
|
|
|
Post by boosters on Sept 29, 2016 17:30:17 GMT
They are, but those aren't configurable in sandbox mode, are they? If so, happy to include those parameters in the rules as well.
|
|
acatalepsy
Junior Member
Not Currently In Space
Posts: 97
|
Post by acatalepsy on Sept 29, 2016 17:50:32 GMT
The problem with this kind of thing is that the AI just isn't bright enough to use ships designed for human control in an even vaguely correct manner. For example, I'm working on a carrier/drone centric design, and it will absolutely get shredded if placed under AI control. Design for this kind of fight is absolutely designing around what the AI knows how to use, not what is actually effective against an opponent who isn't dumb. It also pushes heavily for low-dV fleets, because the AI doesn't maneuver, and won't try to punish you for not having enough dV by dodging your intercept and then laughing as you go on a one way trip to nowhere.
My suspicion is that missiles in particular are overpowered when used against the AI (because they won't dodge or change orbits effectively, or use drones effectively to interdict incoming missiles), and underpowered when used by the AI (because the AI has no idea how to actually use missiles effectively, and good human players can just dodge them; not to mention use 'smart' decoys that the AI has no idea how to prevent). And missiles are cheap enough within this limit that they'll likely be a dominant strategy; every fleet will include a missile truck of some sort and every fleet will wipe every single other fleet (when controlled by a player against an AI). On top of that, I think we're waiting for at least some review of the questionable physicality of certain designs - most notably supereffective micro-NTR rockets and minifission reactors.
Also, uh, without streaming or some kind of verification you can't stop someone from winning every fight by just trying again and again until they get it right, then posting three victory screenshots.
|
|
|
Post by boosters on Sept 29, 2016 17:58:23 GMT
The problem with this kind of thing is that the AI just isn't bright enough to use ships designed for human control in an even vaguely correct manner. For example, I'm working on a carrier/drone centric design, and it will absolutely get shredded if placed under AI control. Design for this kind of fight is absolutely designing around what the AI knows how to use, not what is actually effective against an opponent who isn't dumb. I think that is true currently, but do we even know if proper multiplayer is in the works? Or will we be stuck with play-by-post against questionable AI forever? Yup, I experienced the missile problems last night myself. With a 250km engagement range for my lasers, the incoming missiles just burned straight at me until empty so they were on a ballistic trajectory. I evaded the missiles by boosting forward slightly, lol. So that needs work. Bah, nobody would ever be dishonest on the internet. And if we suspect anyone of cheating we can just type invective at them incredulously.
|
|
acatalepsy
Junior Member
Not Currently In Space
Posts: 97
|
Post by acatalepsy on Sept 29, 2016 18:32:36 GMT
I think that is true currently, but do we even know if proper multiplayer is in the works? Or will we be stuck with play-by-post against questionable AI forever? Probably? Multiplayer is a whole different problem; one filled with fundamental design and coding issues. It's not unsolvable, but it's a major, major undertaking that's out of line with what resources appear to be available. You'd probably want at least one other developer dedicated to that, as well as doing more intensive QA because multiplayer is bug city. I think the only way multiplayer will ever be a thing is if qswitched decides to go all in on this. That would probably mean launching a Kickstarter, or talking to a publisher like Matrix Games (who publish a lot of niche simulation and wargames, such as Command: Modern Air and Naval Operations, or Starshatter). If that happens, I'll certainly be there, but it doesn't seem terribly likely at this point. Yup, I experienced the missile problems last night myself. With a 250km engagement range for my lasers, the incoming missiles just burned straight at me until empty so they were on a ballistic trajectory. I evaded the missiles by boosting forward slightly, lol. So that needs work. Right, the arbitrary laser ranges, combined with the lack of intercept logic for missiles, make missiles basically nonfunctional. With the somewhat-ridiculous 10 kps+ NTR missiles and some careful hand-maneuvering, it may be possible to engage a target starting 250 km away, but it's not exactly fun, easy, or interesting. A more robust approach might be to use heavily laser-armored drones with missile magazines as a sort of multi-stage missile and use raw dV to overcome, but still - it's not terribly great.
|
|
|
Post by boosters on Sept 29, 2016 18:47:15 GMT
I totally agree about the scope of the implementing multiplayer - that sounds like as much work again, potentially, as putting out this game. Though one thing that comes to mind is that qswitched has a forum full of amateur (and possibly professional?) rocket scientists, no doubt with quite a bit of combined software experience. So, volunteer work wouldn't be hard to come by, right? I certainly wouldn't mind committing some lines of code to something like that, though multiplayer gaming would be a new application for me. I would absolutely throw money at that hypothetical Kickstarter, though. Single player is really fun, but to answer the question this game is meant to solve (how do space combat?), I think multiplayer is a must. It's the One Big Feature on my wishlist, the rest just seem like quality of life improvements.
|
|
|
Post by blothorn on Sept 29, 2016 18:56:49 GMT
I am very excited about the possibility of tournaments, but I agree that human-vs-AI is probably unworkable. Specifically: * Autocalculated kinetic ranges are much lower than effective, and AFAIK the AI never uses "ignore weapon range". * It is easy to get AI missiles to engage decoys where a player would probably just scatter--and if not, the AI has weaknesses in handling missile thrust to get hits on maneuvering targets. Put those huge advantages together with more subtle optimizations, and I suspect that most competent players would go 3:0 against everything.
What would work, and would probably take relatively little developer time, is allowing AI-vs-AI battles. That is how almost every tournament is run in the other design game I play (From the Depths), and has worked quite well. (Incidentally, FTD did get workable multiplayer when it was still a one-man dev team, although I think MP has always had enough severe bugs to just teeter on the edge of usability.)
Edit: I also think there are some rough edges in the physics calculations (particularly for reactors) that need to be hammered out.
|
|
|
Post by boosters on Sept 29, 2016 19:08:17 GMT
I am very excited about the possibility of tournaments, but I agree that human-vs-AI is probably unworkable. Specifically: * Autocalculated kinetic ranges are much lower than effective, and AFAIK the AI never uses "ignore weapon range". * It is easy to get AI missiles to engage decoys where a player would probably just scatter--and if not, the AI has weaknesses in handling missile thrust to get hits on maneuvering targets. Put those huge advantages together with more subtle optimizations, and I suspect that most competent players would go 3:0 against everything. What would work, and would probably take relatively little developer time, is allowing AI-vs-AI battles. That is how almost every tournament is run in the other design game I play (From the Depths), and has worked quite well. (Incidentally, FTD did get workable multiplayer when it was still a one-man dev team, although I think MP has always had enough severe bugs to just teeter on the edge of usability.) Edit: I also think there are some rough edges in the physics calculations (particularly for reactors) that need to be hammered out. Wow, I hadn't considered AI-vs-AI "autobattles". That sounds like a promising and less labor-intensive method of testing designs against each other, while we wait around for proper multiplayer. Cool idea!
|
|
acatalepsy
Junior Member
Not Currently In Space
Posts: 97
|
Post by acatalepsy on Sept 29, 2016 19:23:30 GMT
Single player is really fun, but to answer the question this game is meant to solve (how do space combat?) I think multiplayer isn't sufficient, and it probably isn't necessary. I think that there are a lot of hard problems that this game can help solve, or at least raise some very interesting questions about - if nothing else it's an amazing place to refight the old purple-green laser-missile wars. But a full answer to "how do space combat" requires, I think, more than an ability to put ships up against each other in the style of a death-match; it requires a harder look at assumptions of economics and logistics which can't really be done in simulation (or, at least, this simulation isn't set up for it), and must be argued for or demonstrated through citation. Incidentally, there's an easy-enough way to simulate any fight you want for the purpose of answering questions, which is to simply give the player the ability to control both sides in a fight, along with an ability to pause (and, ideally, the ability to save a game so that specific scenarios can be worked through multiple times). No multiplayer required, though a community that challenges assumptions and suggests different approaches - taking each other's saves and showing how to get one side or another to win an engagement - is probably essential. Wow, I hadn't considered AI-vs-AI "autobattles". That sounds like a promising and less labor-intensive method of testing designs against each other, while we wait around for proper multiplayer. Cool idea! That might be an interesting game, but it's a very far cry from answering "how do space combat"; you end up building around the limitations of the AI to the point that the fleets themselves are just gross over-optimized parodies.
|
|
|
Post by boosters on Sept 29, 2016 19:53:46 GMT
But a full answer to "how do space combat" requires, I think, more than an ability to put ships up against each other in the style of a death-match; it requires a harder look at assumptions of economics and logistics which can't really be done in simulation (or, at least, this simulation isn't set up for it), and must be argued for or demonstrated through citation. This is also true - the current game hand-waves that stuff away, which I think is okay, but there's definitely a place for simulating more restrictive economic constraints within an engine that doesn't enforce them well by default. If a strong argument were made for the availability or lack thereof of certain materials, despite their solar system abundance, then a modder or developer could (and should?) go in and change the credit costs, I think. This sounds like a great idea, and as you say, far easier than multiplayer. Towards this end, it seems like development time could be better spent improving our ability to manually control our fleets (better commands, shortcuts, maybe some automation), but I think this has a lot of promise, especially the saved scenario part.
|
|
|
Post by boomertiro on Sept 29, 2016 20:01:25 GMT
Considering the issues, making the tournament AI vs AI sounds good to me.
|
|
|
Post by blothorn on Sept 29, 2016 20:28:56 GMT
I agree that AI vs AI is suboptimal---but I think it much better than human vs AI. The former involves working around the AI's limitations, the latter exploiting them. A lot of strategies I have seen only really work because the AI is stupid (decoying missiles with low-value targets) or does not exploit the engine fully (in particular, not ignoring range).
I would like to see a scriptable AI--it would be a fair bit of work, but would allow crowdsourcing the (far harder, I think) problem of actually developing good AI.
|
|
|
Post by quarkster on Oct 1, 2016 18:41:37 GMT
I also prefer AI vs AI
|
|
|
Post by wazzledazzle on Oct 2, 2016 15:27:16 GMT
AI vs AI makes sense. By the way, I had thought about this a little. CoaDE is excellent for tactical battles, but lacks a strategic layer. Meanwhile, the board game Battlefleet Mars has a grand campaign that spans across the entire solar system, but battles are resolved with a simple dice roll. Oh, and it has a hard-fi theme as well. So, if we can get asynchronous battles working, that could be an idea for a forum game. Use BFM for the strategic part and resolve battles with CoaDE ?
|
|