Loving this mod. Had a pretty epic match on 5 small moons with 5 AIs in FFA. One AI preceeded to kill all the rest and then nuke the crap out of me. It put up a a great defense, but it had some bug with halleys. It only ever built 1/2 needed for some reason. If it had not done that it would have won for sure.
You got a link to that game so I can check it out? I haven't done anything with halley's yet, so it probably just couldn't afford to build the remaining ones. That's especially likely if it was pouring resources into nukes.
Version 0.4.1 Reverted AI land platoon sizes to vanilla settings as it seems the AI is better with large numbers of small platoons than small numbers of large platoons Known issues Commander sometimes sits there doing nothing, primarily on maps with dense terrain Fabbers sometimes sit there doing nothing First fabber always assists as its first action (CANTFIX) AI doesn't seem particularly bothered about defending its base AI will not expand via teleporters, it has to use transports (CANTFIX) Scouting units won't reliably attack targets of opportunity (CANTFIX) AI will often take much longer than necessary routes to attack (CANTFIX) Fabbers don't always take the closest metal spot AI doesn't defend its fabbers AI dithers with units around buildings sometimes instead of taking them out (CANTFIX) AI is based around a 1.0 economic multiplier and doesn't ultilise eco boosts effectively
I'm actually very impressed by this. I look forward to seeing some more stuff I really hope Sorian can fix the unit dither bug...
Upcoming AI changes that should help AI mod efforts Fixes/updates: All of the known issues of the AI stopping and doing nothing should be fixed. The AI should no longer build over metal spots, unless the is a large density of metal on the planet. Added adv_eco_mod to AI personality, which controls how much eco the AI needs before going to Advanced. Split AI personality percent_land into percent_vehicle and percent_bot (percent_land is still supported). AI starting factory will be the highest of percent_vehicle, percent_bot, or percent_air. Platoon build item instance counts now work. New build conditions: AllMetalSpotsFull - True if no base has an empty metal spot. UnableToExpand - True if the AI is unable to expand (create a new base) due to no room or too much threat. The above are currently used to detect that the AI has no more room to expand and needs to tech up in order to keep expanding its economy. NeedBasicVehicleFabber NeedAdvancedVehicleFabber NeedBasicBotFabber NeedAdvancedBotFabber NeedBasicVehicleFactory NeedAdvancedVehicleFactory NeedBasicBotFactory NeedAdvancedBotFactory Removed the Need*LandFabber and Need*LandFactory build conditions The above changed due to the AI personality changes. There are obviously a lot more changes then this, but these are the ones that are likely to be of interest to AI modders.
This is the only change I'm unsure of as it feels too limiting for AI openings which I'd currently handle by going "What did I build first?" and then cut down the build order based on this. Is this always, or only when Need___ is called? Or will there be a way to call random personalities so one weighting bots or vehicles gets called and then the brain runs with it?
Only affects the Need*Factory calls. Feel free to have your own, more build order restrictive, build items. I plan to eventually have a way of letting the AI have a personality where it can adjust based on game state.
wow the ai is mod-able? i should have looked here sooner! question 1 : how far does it go? my biggest need would be to alter the behaviour of the commander (aka if enemy army approaches RUN FOR YOUR LIFE instead of running out of base trying to defend it and being destroyed in secounds...) but i fear the neral network has to be altered for it. i made a quick look over the files and found a big endless stream of numbers. question 2: is anything known about the format of the neural net yet?^^" question 3 : why focus on 1 vs 1 battles? am i the last player that has the original idea of large number of players in mind? question 4 : you said you alter it so it makes more use of teleporters, does this include the ability of nice planetary invasion (to eliminate the boring nuke shootouts when its a multi-planet battle with the ai...)
There's no neural net for the Commander yet AFAIK. Right now your only option here is TransportToPlanet to have him run away to another world in an Astraeus. The default AI has this behaviour. I don't believe there's a way to force him to teleport to another planet. The files aren't designed to be human-readable or modifiable. The parts of the neural net which matter are not under our control. Because it's what I enjoy. Because it's a smaller goal with clear focus. Because it keeps testing shorter. Because it makes more sense to focus on this and expand a strong planet game out to other planets. Because the AI is missing critical elements of the orbital game (coming in the next patch). There's a TeleportLandToPlanet which is used by the planet of origin. You can't control how the AI hooks up its teleporters though. My focus is on making sure it has teleporters in the first place and that it builds one first when landing on another world to maximise the chance of a successful invasion.
1. i didnt mean him to teleport to another planet, i just meant when his base gets attacked from south, he runs north to be out of shooting range instead of walking south to defend his base. but since you mention it : the transport to other planet is also a big problem, for example when there is moon in game, he tends to send the commander there, sadly if more than 1 ai, sometimes to the same spot the same time, so one commander shots down the astareus with other commander etc...but ok this is not relevant for 1 vs 1 except you are on the moon first and he tries to land exactly on your com.... 2. yes i see that its not modable "by hand", but knowing the format would be nice so you can write a small c program or whatever to read in, modify and write out again. thou when the relevant parts of neural net are unaccessable this mostly makes no sense then.... 4. i have seen ai build teleporter first, but what i didnt see till now was that he sends a big army in the teleporter to invade. is this maybe possible to teach him?
Version 0.4.2 After reverting platoons I forgot to revert the squad percentage makeup to my own. Now fixed. Known issues Commander sometimes sits there doing nothing, primarily on maps with dense terrain Fabbers sometimes sit there doing nothing First fabber always assists as its first action (CANTFIX) AI doesn't seem particularly bothered about defending its base AI will not expand via teleporters, it has to use transports (CANTFIX) Scouting units won't reliably attack targets of opportunity (CANTFIX) AI will often take much longer than necessary routes to attack (CANTFIX) Fabbers don't always take the closest metal spot AI doesn't defend its fabbers AI dithers with units around buildings sometimes instead of taking them out (CANTFIX) AI is based around a 1.0 economic multiplier and doesn't utilise eco boosts effectively
Currently working on v0.5, which will include some variation in the openings and what I hope to be a half-decent naval game. Here's a video of a 1v2 test game with my current development version on Systeem X. All things considered I think Queller did pretty well, especially given one of the AIs ends up right on top of me which is much tougher on the AI than it is on me. Orange holds equal with me on eco for almost the entire game. Certainly some kinks to work out, like that laser cluster, and the fact that blue went heavy air when really being that close means it needs to switch straight to tanks. I need to look at better mid-game factory expansion and stop it investing too much in air if it's not winning on the ground. Bomber fleets looked impressive though, I think it could have won with a snipe. Sorian is making some changes to the AI's code that means next time it might just do that.
They work will at harrassing early game, and actually bring somewhat of a challenge early game... But only early game. They literally never seem to touch Vehicles allowing me to go in and swoop them clean. It beats Stock AI in early game but seems to be lacking factories around the 5 minute mark. where the stock AI shines and begins to put up a fight with ants.
My experience is that Queller will have more units than stock at every phase of the game when everything works. The current release version will build a number of bot factories based on the size of the planet, then stop building them and rely purely on vehicles and air. You'll see this behaviour in my 0.4.1 test video where it starts out bots and transitions to heavy vehicles. It's very reliant on the commander though for its factories. I've found some maps which seem to cause the commander to stop building, which in turn will cripple Queller. Once I've experimented with this some more I'll report it to sorian. What systems are you playing on? It'd be handy if you check back on games you play and see what the real mobile count was so you can check whether stock really did have more units than Queller, or it just felt like it.
Another test of v0.5-dev, this time with naval on Waterworld from the Team Genesis map pack. Warning, I take a long time to pick my spawn. Overall I was relatively happy, the AI did what it was supposed to. Need to look at tightening up that eco handling a bit, it can't rely on the same techniques it uses on land keeping the economy running. Also need to stop it building those useless laser turrets on the water instead of the far more useful torpedo launcher. Clearly I need to work on my naval game, it will be hard to improve the AI here if I don't know what I'm doing myself. It didn't help that I was pumping out Narwhals when I wanted Orcas.
