The new guy: A.I. Developer #42 & Discussion on A.I.

[Dix]

After all i've read in this thread, i'm really looking forward to fearing the uncannily clever Ai Reaper. All I want to see from the ai now is gatherer/herder missions - aliens loading civvies in to their ufos at gunpoint. To be stored and then fed to the chrysalids . (Or wait, are chryssalids from a differant game . I really need to download and play this thing, but I don't want to spoil the gooey chocolatey goodness of firing up the game for the first time and playing it all the way through, as Chris intended.)

[StellarRat]

What I'm scared of is doing everything "right" (good cover, overwatch, etc...) and still being overrun by aliens that my weapons simply can't stop in time. That coupled with a good AI is truly frightening prospect. That's what I hated about Lobstermen and Chrysalids in the old games. Sometimes you simply couldn't get enough firepower on them to stop them even if they basically charged into a virtual firing squad of your troops. If the new aliens use cover to advance before the final "charge" it's going to be a nightmare! At that point I'll have to consider whether or not to even go on some missions without bigger, better weapons.

[Jean-Luc]

I also talked to the development team of XCOM:EU, and point is pretty simple: They're aiming for a different target group, instead of the traditional X-com fans. This is just one of the byproducts of being forced to ensure your product can return on the (large) investments.

When did you get a chance to talk to them? In what sort of context? A private one maybe?

[Gijs-Jan]

When did you get a chance to talk to them? In what sort of context? A private one maybe?

At the last GamesCon in Aachen, I was allowed a private one-on-one with two members of the team. The debate essentially ended in a (very) friendly us-vs-them of XCOM:EU and Xenonauts. I really got the feeling that even some of the higher-ups in the development team were kinda disappointed with some of the streamlining in the game. (The amount of guidance/tutorial missions was ridiculous in the showcased build)

On the other hand, some of the streamlining did make some sense.

From my limited perspective, I would say the fact I was disappointed with was the amount of expected re-playability. Levels are fully preset, localized to the region of the mission. In one play you shouldn't see any duplicate levels, but in replays, you will. The attention to detail was tremendous though; per example, if you would play in Germany, you would see locally textured police cars.

The game feels heavily choreographed in presenting a thought-out, polished story. I wasn't able to play the game at higher difficulties, so I don't know how the difficulty scales up.

And now I'm contributing to derailing this thread!

A small update then: I've been primarily busy the last few weeks with further designing of the system to be implemented. For those interested of things to come, some of the details of the system could be compared with articles found in the Game AI Wisdom series. I'll try to get another blog post up in a week.

[Jean-Luc]

Thanks GJ, just what I wanted to know.

The game feels heavily choreographed in presenting a thought-out, polished story. I wasn't able to play the game at higher difficulties, so I don't know how the difficulty scales up.

It's hard to tell from the videos but I got the impression that the progression in Fix-Com very much resembles a standard mission-by-mission campaign with a tightly scripted series of events and that it lacks the sandboxy style of the original.

I feel it would've been quite a different game if the devs just had their way with it without having to cater to commercial interests. But I don't blame them.

[LeftyRighty]

It's hard to tell from the videos but I got the impression that the progression in Fix-Com very much resembles a standard mission-by-mission campaign with a tightly scripted series of events and that it lacks the sandboxy style of the original.

The TUTORIAL is a sequence of scripted events, and was given out to the press for the initial hands on. So a lot of the gameplay reviews as of a month or so ago have been based on that. Which, I'd agree, probably give a bad impression of the gameplay. It was also the handson version that you could play if you attended the recent gaming events like PAX and GamesCom.

The tutorial is completely optional. Outside of that the game is random ufos to be shot down yielding ufo missions, terror sites and they've hinted at but not explicitly confirmed alien bases. They've said on several occasions that the AIs used for the alien actions on the strategic and tactical layers are independent of player actions, and their options are only limited by the difficulty setting. So it is the same sandbox model.

Recent screen shots and vids have shown there are "objectives", for example the game tells you at one point to research and build a stun weapon to capture a live alien. This is similar to the original game requiring you to capture a commander and build an avenger to progress to cydonia, except instead of burying it in the research descriptions it's explicitly stated (and given the modern games culture it's a familiar mechanic for general gamers).

They really need to released more about the strategic gameplay, they've concentrated far too much on the tactical for the press release and sneak peeks... it'd probably help with the impressions of gameplay.

Edited September 12, 2012 by LeftyRighty
wasn't the tone i was aiming for, added clarity

[XenoMask]

Yeah, Firaxis XCOM: Enemy Unknown works something like this: There are few story missions which are probably set(maybe with random enemy/starting placement, but map will probably be same), but every other mission picks map from map pool, then randomly decides starter location and enemy placement and what enemies there are

[Max_Caine]

LeftyRighty, while there might be alien bases, according to this penny arcade report there will be no alien base invasions (right at the bottom of the article).

[Jean-Luc]

LeftyRighty, while there might be alien bases, according to this penny arcade report there will be no alien base invasions (right at the bottom of the article).

Is he referring to alien base invasion or alien base invasion. Pretty sure it's the first thing meaning there'll be no attacks on the (one and only) X-Com base.

In any case good to know the campaign won't be so tightly funnelled.

Edited September 12, 2012 by Jean-Luc

[Aleksander]

Hello, Gijs-Jan!

I worked once on xenonauts AI.

I'd like to see good AI in xenonauts, it can increase chances for better AI in all gamedev in future, so I wish your work would be succsessfull. But I am not as optimistic as the community is about getting good AI in xenonauts at last. I believe it is nearly impossible at current stage of computer science to make AI in a group, which would make better decisions then the best expert of this group. Also, as I understand, alien team generally would have strong differences from xenonaut team, due to specific types of aliens which are not like a man with a rifle. So I am intrested - is it planned to make an improvement (maybe it is already made, but I think it's unlikely) of general xenonaut engine to allow human rule aliens against xenonauts (ruled by human or AI) or you are supposed to make AI for alien types, which you won't be able to try first-handed in battle against xenonaut team?

[Chris]

We're not making the alien team a seperate playable entity to allow a human player to play as the aliens to test the tactics the aliens will use first-hand, but I'm pretty sure that won't be necessary. The AI seems to be taking shape fairly well so far, so I'm sure it'll continue to develop well in the future too.

[StellarRat]

We're not making the alien team a seperate playable entity to allow a human player to play as the aliens to test the tactics the aliens will use first-hand, but I'm pretty sure that won't be necessary. The AI seems to be taking shape fairly well so far, so I'm sure it'll continue to develop well in the future too.

Is there any possibility of getting the AI code split out so users can make improvements after the game comes out? If the AI were contained in a separate DLL (with source code available) or there was a API to allow map view and control of the alien movements and firing the user community could mod this.

[LeftyRighty]

from the blog linked in OP:

The implementation requirements presented are:

The A.I. system in place needs to be moddable by the community; a Hierarchical Finite State Machine (HFSM) which is editable through XML is preferred.

[StellarRat]

from the blog linked in OP:

Doesn't tell me too much. If have an good API or the ability to write your own code you can do anything, if you can just change some of the parameters and constants you don't have nearly as much control.

[Gijs-Jan]

Doesn't tell me too much. If have an good API or the ability to write your own code you can do anything, if you can just change some of the parameters and constants you don't have nearly as much control.

What do you want (/expect) to implement, or extend?

[StellarRat]

What do you want (/expect) to implement, or extend?

I'm not sure, I haven't seen your finished product yet. :-) But, it would be nice if we could have an API to read and update target prioritization values, weapon use, how much AP the aliens reserve, read terrain types (i.e. the map), read and update terrain valuation (i.e. the tile values), read aliens locations, read alien type and weapons they carry, and get a picture of what each alien "knows"/"can see" each turn. If those things could be accessed through code or scripting it would be possible for someone to change or fix nearly anything they don't like about the AI. I don't think pathfinding would need an interface if your implementation is good (I'm sure it will be.)

I'm not saying you're going to do a bad job, but someday you will no longer be working on the AI and I know there no such thing as a perfect AI, so there is always room for improvement. Obviously, if it's a lot of work to add some code hooks, then I don't expect it to happen. Just something to consider.

Also, if Goldhawk thrives and needs more programmers you might find some unknown AI talent in one of the modders. It's happened before. Then again maybe no one will attempt to extend the AI, you never know.

Edited September 13, 2012 by StellarRat

[Moonshine Fox]

example of cheating AI

supreme commander on highest level

It only "cheats" in the sense that it has x1,5 resources and buildspeed. It also has Omni sensors on everything, instead of Visual/RADAR/SONAR. It's not omnipotent as your run-of-the-mill cheating AI is. I'd rather call it "AI with handicap" rather than outright cheating.

EDIT: Baah, derail! Sorry >_<

Edited September 18, 2012 by Moonshine Fox

[anotherdevil]

Welcome to the team G-J! Looking forward to getting thrashed in the BETA!

[Max_Caine]

GJ, there's one thing I've had a think about. When you're determining the order in which aliens and civies move, could you try and make it so that the civvies and aliens that the player can see move and act first? that way the screen doesn't flip-flop between the hidden movement screen and the map too much.

Edited September 30, 2012 by Max_Caine

[Gauddlike]

That only works until someone who was out of sight moves into sight.

The screen would still have to jump back to that target at that point.

Agreed though that it would make things less jumpy if all visible movement was done before hidden movement where possible.

Not really an AI matter though I would have thought.

[Gijs-Jan]

Supreme Commander AI

Funny thing though; try placing the AI on a map with only islands...

AI Moddability

The problem with this is that I'm trying to move away from the traditional "If you see Unit X with weapon Y, and Health < Z; then do A, else B". Essentially, I'm trying to move to a system in which we give the aliens a set of actions that are possible, some way to give a score to each action based on some preference, and let the aliens try to calculate what the best action would be. (Instead of explicitly telling it what to do)

Tl;DR: The API will probably be that you can adjust a (long) list of preferences of each alien on how it scores different actions. Depending on how well this works, I'll extend it on community preference. :-)

(The only thing I'm not sure now about is maybe allowing actions to be read from LUA -> performance goes out the drain)

GJ, there's one thing I've had a thought about. When you're determining the order in which aliens and civies move, could you try and make it so that the civvies and aliens that the player can see move and act first? that way the screen doesn't flip-flop between the hidden movement screen and the map too much.

Although I'm looking into reducing the time of the "Hidden Movement" phase, I don't know if this is feasible, given performance constraints. The problem here is than you have a sorting problem with data that needs intensive checking. (LOS)

Hmmm, and while typing this, I think there is a shortcut for achieving similar behavior. I'll try to look into it, and if you don't see it in the Beta, just bump it here again :-). At the very least I'll give an explanation on why I did not implement it.

On the other hand, it seems like a gameplay alteration request, so I'll have to run it by Chris first. At the very least I should be able to "clump" together the aliens with visible movement.

[Max_Caine]

(Written in response to Gauddlike's post)

I would have thought determining which alien/civilian moves first would be an ai matter, as movement and action options for aliens/civvies begin to disappear as other aliens/civvies move and act (especially when is comes to finite resources like available cover).

[Gauddlike]

That would appear to be a game optimisation decision, rather than something that needs to be decided by the individual aliens.

It isn't that the aliens who are in sight have decided that they need to move because they are in more danger, it is that aliens in sight should move first to make the hidden movement phase flow better.

That was why I suggested it wasn't really an AI matter.

[Gijs-Jan]

It would fall into the category of game optimization; yes, but it would have to be done by me.

Quite simply because I would need to track and sort all planned actions on visibility before execution; which could be done.

[Phydeaux]

What about the strategic layer? Will time be spent there? Deciding where to build a base to respond to the Xenonauts threat, setting up supply-lines, searching for Xenonaut bases, and more interestingly: Deciding on force compositions for various missions. If the aliens are sending a ship to attack a human base, they will want to wipe it out. If they know what the humans normally field and what they seem to be ineffective against, it would only stand to reason they would send more of those.