Recommend
1 
 Thumb up
 Hide
20 Posts

Chronicle X» Forums » General

Subject: Brainstorming - Tackling Alien AI for Solo/Coop play rss

Your Tags: Add tags
Popular Tags: [View All]
Thrawn 007
msg tools
Creating compelling AI in a board game is a very interesting topic, and it is very challenging. For a game designed for overlord play, adding in a workable AI can be even more challenging. As a huge fan of Coop games, I want to make sure this piece of the game is a viable part of the game and not just and afterthought. I know many others are interested in solo play, so there is definitely an audience specifically for this part of the game.

Usually, coop games are designed around players vs the system. A lot of the time, the "game" doesn't have that many decisions to make, and the challenges it presents are all baked in, so no AI system is needed. Dungeon crawlers without an overlord usually only need rules for attacking, as they are either randomly placed into battles, or are part of pre-scripted scenarios.

Chronicle X brings to it a number of challenges when creating a workable AI system.
Challenge 1: Force selection
Forces are preselected at the beginning of the game. So not only does the system have to include a way to select units, it needs to be able to react to having different mixes of units and still be able to accomplish tasks.

Challenge 2: Blips
Blips are great in that they give an unknown to the players, but how should blips react? Is the support unit charging up front at the heroes? Is the killing machine the one moving off to try and hack the computer? We don't even know because they are still blips.

Challenge 3: Accomplishing Missions
We have a number of different missions for the aliens to accomplish. If the only mission was killing all heroes, then AI would be a much easier task, as we could just set everything to the most efficient way of destroying things. However, often the aliens will be tasked with different objectives. What units should try objectives? What units should protect objectives from heroes? What units should just go kill things? We don't know all the objectives yet, but it is very likely some types of aliens (which we may or may not even have in our current force) are more adept at objectives, and others will be better at fighting.

Challenge 4: Combat
I see this as the easiest challenge. (Selecting forces can actually be easy, but adapting tactics based on that selection isn't.) However, there are things that shouldn't be trivialized. How will aliens determine threat/aggro to pick targets? Will aliens have multiple options of attacks? If so, how do they choose which one?

Getting these areas right, can give a game that is a fun puzzle solving game with a lot of variables. Getting these wrong will make the coop mode come off flat, and simply be an inferior and random substitute for a player overmind. AI will almost never play as well as an overlord, so it's likely some kinds of "cheats" will need to be introduced to balance things and help the overmind out.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
The best AI I have seen in any game is in Myth. That was one of the two things that attracted me to that game. (The other was the play of hero decks.) Myth only had to deal with challenge 4 out of our challenges in CX, but it did that well. However, the entire game was built around how the AI would work.

It solved a number of things around combat in elegant ways:
1) Players generated "threat" asa they did actions. This gave one of the components of who enemies should attack.
2) Standard monster AI was based on ranged or melee. Melee attackers always moved towards their selected target. Ranged attackers would try to stay at maximum range and shoot at their selected target.
3) Target selection was the main behavior of each creature. They had about a 4 item priority list of preferred targets based on things like target health, character type, proximity, or threat rating. They would find a valid target based on their top priority (such as proximity) going to their second priority (such as lowest health), and only go to the third if needed (such as arcane casters). Different creature types all had different priorities. Many just attacked who was most convenient, while others would try to concentrate fire and take out certain types of targets. This mix is what makes the fights come out different.
4) Creatures had base attacks, but "Fate dice" allowed the use of extra abilities. Crreatures rolled 6 sided dice that had different symbols on them. If the symbols match a recipe, some special effect would go off. They fill recipes from top to bottom of a list on the monster's card, removing dice as they fill the recipe. This means creatures with a lot of fate dice might actually kick off multiple recipes in a turn if they get lucky.

These 4 factors created a fairly quick, but very diverse system. Sure, players could game some things like playing an MMO. "If I do this, I'll get more threat, and all the bugs will attack me...so you will be safe" but that is all part of the strategy. Making decisions to impact play was far preferable to all things being random. Of course, there is definitely randomness too, and when some of the special abilities kick in, someone might regret taking all those attacks.

-------------------
Now, in CX, as I said, even if the game was built around systems like this (which it most definitely isn't), this would only solve 1 of 4 issues. Players picked their own enemies in Myth, building the game as they went (at least in free play), so it didn't have army building issues. You really only played a single tile and skirmish at a time. Monster's showed up on board, so we didn't have to worry about stealth movement with blips. And Myth monsters' only duty was to kill heroes. Objectives were completely beyond them.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
Using Active Intelligence instead of Artificial Intelligence:

The likely solution to this is that player input/interpretation will be required to run the aliens.

An AI card may give some guidelines for overall strategy of the aliens for the turn, but it will be up to the players to determine how the aliens will execute under those guidelines.

Pros:
Requires far less complexity in alien design.
Removes some of the randomness that can create very suboptimal effects.
Takes into account most of the 4 challenges.

Cons:
It becomes players playing against themselves instead of against the game.
Doesn't solve some optimization play like blip movement
Interpretations of how to execute moves can vary widely causing disagreement.
Gives more room to "cheat" (although if you are cheating at a game, especially a solo game, you have issues.)


 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Silver Bowen
United States
Austin
Texas
flag msg tools
Woo! Gonna get some
badge
44, that's me!
Avatar
mbmbmbmbmb
Thrawn007 wrote:
Using Active Intelligence instead of Artificial Intelligence:

The likely solution to this is that player input/interpretation will be required to run the aliens.


As a long-time wargamer and soloist, I've played many 2+ player games solo by simply playing each side to the best of my ability. While often fun, it is no substitute for a hands-off AI and absolutely should not be advertised as a solo mode. The only player decision space in a true solo mode should be among multiple equivalent or near-equivalent options, and then it should be clearly spelled out that the player can or can't take the best/worst choice. For instance, D&D Adventure System games encourage the player to place monsters in the most advantageous (for the player) spot, while Galaxy Defenders says the opposite. Neither leave more than a little bit of room for this sort of thing, though.

A hands off solo mode is much more difficult to develop, and generally requires flowcharting and/or components like card decks to indicate pathing/priorities. This sort of mode is my primary reason for backing Chronicle X, and personally I'd be very disappointed in a player-decision driven mode instead.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Silver Bowen
United States
Austin
Texas
flag msg tools
Woo! Gonna get some
badge
44, that's me!
Avatar
mbmbmbmbmb
Note: I'm assuming Thrawn 007 is one of the designers. Please correct me if that isn't the case
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Leif Stensson
msg tools
mbmbmbmbmb
Thrawn007 wrote:
And Myth monsters' only duty was to kill heroes. Objectives were completely beyond them.

I haven't played Myth, but from your description, it sounds fairly similar to Zombicide. You even generate threat in Zombicide, except it's called "noise". (The target selection bit was missing in Zombicide, though. The zombies just choose to attack a zone, and if there was more than one player in the zone, the players chose how to split any damage between them.)

But as you say, the differences about army building and having some kind of objectives means it will be difficult to use that kind of approach for CX.

For army selection, perhaps you could reserve a corner of the mission card to say what the AI alien's basic army should look like, and then build the rest of the army by just cyclically adding more of the units listed as long as there are spawn points left. For example: let's say the card says 2 GRAY COMMANDERS, 2 DEVOLI RIDGEWALKER, 3 GRAY INFANTRY. We don't yet know what the spawn points of those are (or has that been mentioned anywhere?), but let's for the sake of the example say that the cost is 3 for commanders, 2 for ridgewalkers, and 1 for infantry. The forces listed then add up to 2*3+2*2+3*1 = 13 points. Let's say we're supposed to spawn 20, so first add one more commander, bringing the total to 16, then add one ridgewalker, bringing the total to 18, then add one infantry, bringing the total to 19. Then try to add another commander, except it would cost 3 points, and we only have 1 left, so skip that. Then try to add another ridgewalker, but that would cost 2, so skip that as well. Then add one gray infantry for 1, for a total of 20.

This wouldn't require any extra game components, just a few extra paragraphs in the rules, and a few extra words or icons on the mission cards.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Leif Stensson
msg tools
mbmbmbmbmb
silverbowen wrote:
Note: I'm assuming Thrawn 007 is one of the designers. Please correct me if that isn't the case
He isn't, but some backers have suggested he should be.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
silverbowen wrote:
Note: I'm assuming Thrawn 007 is one of the designers. Please correct me if that isn't the case


I'm not. I'm a backer with a lot of design and playtesting experience, and I'm looking at how difficult the AI can be for this game. A lot of people are looking for immediate answers on how the AI will work. I think it's going to be very complicated to create and test an AI that is the proper mix of programmed (which gives a better experience much of the time) and random (which gives a more replayable experience.)

I'm greedy, and want both satisfying and replay value, so I'm looking at ways we can get there.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
lstensson wrote:
Thrawn007 wrote:
And Myth monsters' only duty was to kill heroes. Objectives were completely beyond them.

I haven't played Myth, but from your description, it sounds fairly similar to Zombicide. You even generate threat in Zombicide, except it's called "noise". (The target selection bit was missing in Zombicide, though. The zombies just choose to attack a zone, and if there was more than one player in the zone, the players chose how to split any damage between them.)


Threat is very similar to noise in Zombiecide. Different actions increase threat in Myth, much like different actions raise noise in Zombiecide. (Myth has the added element that as you build up threat, it determines when "the darkness" gets a turn as well, so the more you generate, the more often the enemies get to act.)

Quote:

But as you say, the differences about army building and having some kind of objectives means it will be difficult to use that kind of approach for CX.


I think making an AI that can successfully balance going after objectives and going after enemies is the key element. Everything else is small compared to this.

Quote:

For army selection, perhaps you could reserve a corner of the mission card to say what the AI alien's basic army should look like, and then build the rest of the army by just cyclically adding more of the units listed as long as there are spawn points left. For example: let's say the card says 2 GRAY COMMANDERS, 2 DEVOLI RIDGEWALKER, 3 GRAY INFANTRY. We don't yet know what the spawn points of those are (or has that been mentioned anywhere?), but let's for the sake of the example say that the cost is 3 for commanders, 2 for ridgewalkers, and 1 for infantry. The forces listed then add up to 2*3+2*2+3*1 = 13 points. Let's say we're supposed to spawn 20, so first add one more commander, bringing the total to 16, then add one ridgewalker, bringing the total to 18, then add one infantry, bringing the total to 19. Then try to add another commander, except it would cost 3 points, and we only have 1 left, so skip that. Then try to add another ridgewalker, but that would cost 2, so skip that as well. Then add one gray infantry for 1, for a total of 20.

This wouldn't require any extra game components, just a few extra paragraphs in the rules, and a few extra words or icons on the mission cards.


I also buried a coop army building technique in my missions thread. (It's in the rules outline post.) It's slower than what you suggest, but I like the versatility in dealing with expansions. One of the biggest concerns with army building is making sure the following:
1) It works with the based box set with stretch goals or add-ons
2) It works with all the stretch goals and add-ons
3) It works in the market, when someone purchases the game, and adds on piecemeal.

In all 3 cases, it should not give results that can't be supported (requiring models that someone doesn't own), while at the same time, not overly limiting the results (only one mission in the game uses a particular model, so you almost never get to use the expansion you spent good money for.)

Scripted campaign modes don't have to worry about this. But I think this game will be better suited to a more free flowing army comp set of rules.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Leif Stensson
msg tools
mbmbmbmbmb
Thrawn007 wrote:
I also buried a coop army building technique in my missions thread. (It's in the rules outline post.) It's slower than what you suggest, but I like the versatility in dealing with expansions.

OK, found it. I'll quote and comment it here rather than in the other thread, since this thread is more specifically about the AI:

Thrawn007 wrote:
Solo/Coop Mode: Either the heroes can create an alien army using the rules above, or step 3 can be replaced by the following:
4) Select a random composition for the alien army by following these steps.
a. Take the cards for all available unit types (as determined in step 3) and shuffle them together.
b. Draw the first card from the top of this deck, adding the selected unit to the overmind’s forces, and deducting the appropriate number of points from the maximum force value.
c. If there are still available models for this unit type and the cost of this unit isn’t greater than the remaining points, shuffle the card back into the deck.
d. Repeat steps B and C until no more points remain to purchase units.
e. Repeat A through D for special point pools, except use only a deck of cards available for the special point pool instead of the full list of available units.

I feel that approach has two shortcomings that both are consequences of "this isn't AI, it's just random":

- The set-up becomes unrelated to the mission, so you can get a much easier mission if the random army building picks units that aren't adapted to the mission. And you'll get silly stuff like underwater missions with just land minis, and then the underwater creatures show up in a mission that storywise takes place in a desert.

- The various units making up the army may not use any synergies or other cooperation advantages, which in addition to making the aliens seem kind of stupid also makes them weaker than their point value suggests.

I think some of the alien's forces need be predetermined for the solo mode by the mission to avoid that. You could add random selection of the additional forces, but in situations where there are many spawn points left after the predetermined starting forces, it will tend to tip back to the "more random than AI" situation.

Maybe a reasonable compromise could be to start with the predetermined list, then randomly select ONE other mini (if there are enough spawn points for it), and then top up the forces with more copies of the units already included, including the randomly selected one?

It would be nice if there was a way of making this really flexible at the same time as it's easy and makes sense, but I think that risks making the system rather complex both to playtest and to perform the setup for each time you play.

Thrawn007 wrote:
One of the biggest concerns with army building is making sure the following:
1) It works with the based box set with stretch goals or add-ons
2) It works with all the stretch goals and add-ons
3) It works in the market, when someone purchases the game, and adds on piecemeal.

Agreed, those are all desirable to achieve. But if add-ons include extra mission cards using the new minis, that goes at least some way towards addressing that. Then, you could decide to separate the mission cards by expansion, and randomly remove half the base game's mission cards when you're playing with an expansion, to avoid having too many missions that don't use the expansion's minis.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Leif Stensson
msg tools
mbmbmbmbmb
New info from Glenn Allen in the Q&A session: the current plan for the AI mode is that there's a different story. The world map and the whole reclaim Earth story is out, and instead, it's the human space station that is "losing energy" and needs to be repaired, and the missions will apparently be centered about that idea. Componentwise, there will be some kind of score/progress tracker to replace the world map when playing story mode.

This doesn't make any huge difference for what we've suggested so far, but it does change the story/flavour of the game a bit, which might affect how army building should be done.

---
EDIT: another related bit of info from Glenn: the human mission cards will probably be the same for solo games as in versus games, but the alien mission cards will probably be different.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
Yes, the solution to one of the challenges (which was going to be the hardest to solve) was to remove the missions from the aliens. I can live with that solution, because I know for me, I couldn't come up with really good answers to solve that one either.

For other challenges:
Blips would be move using AI cards.
In combat, models will work based on programmed behaviors.

Seems like a pretty straightforward solution to those challenges.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Bryce K. Nielsen
United States
Elk Ridge
Utah
flag msg tools
Avatar
mbmbmbmbmb
Thrawn007 wrote:
The best AI I have seen in any game is in Myth.

How does it compare to Gears of War: The Board Game AI deck? I found GoW to be the best in my experience, but I haven't actually played Myth to know how the two compare...

-shnar
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
We have no idea yet. We know there is a deck, and that it will control blips. Outside of that, details are still being worked out. It's not even in alpha stage yet, just conceptual stage.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Leif Stensson
msg tools
mbmbmbmbmb
Thrawn007 wrote:
We have no idea yet. We know there is a deck, and that it will control blips. Outside of that, details are still being worked out. It's not even in alpha stage yet, just conceptual stage.

No, I'm pretty sure he meant how the AI in Myth compares against the AI in Gears of War. (I haven't played either of those games, so unfortunately, I can't answer that question.)
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
Gears of War is a set of cards with if/then statements. It's actually what I was kind of expecting the AI for this game to look like before the Q&A. This system creates a lot more randomness in the AI than a more programmed AI, but it can still be designed to allow for logic to it.

Because the instructions are on the card specific to units, it has a problem with scalability when you are looking to activate all units.

Myth by comparison is far less random (but includes random elements) and is more scalable.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Jeff Fike
United States
Minnesota
flag msg tools
mbmbmbmb
The AI in myth was awful.

The one credit I give is the idea of threat being generated. But it was easy for players to "Game the AI" and cause threat on one person instead of another. It was a rubbish AI system.

Gloomhaven and Sword and Sorcery had far superior AI systems.

Gloomhaven was best with the concept of initiative and the AI being extremely unpredictable. The issue with a single "ranged AI" or "Melee AI" (Sword and sorcery had this problem as well) is you can predict, to the letter, exactly what the AI will do when it takes its turn. Not the case in Gloomhaven.

The initiative system of Gloomhaven was brilliant. You also could not predict who would go first or in what order.

The drawback to Gloomhaven AI is that some AI cards would result in teh enemy healing itself when it was already at full health, effectively doing nothing. Most would argue the difficulty was balanced well enough to account for it, but those moments did make it look weak.

The Redjak variants for Descent and Imperial Assault are another good example of AI. Again, having the same issue of knowing what the AI will do on its turn, but the way event cards are drawn between rounds and then the AI captures priority targets is really well done.

Sword and Sorcery AI is similar. Priority targeting based on parameters which may change who is targeted.

For me, the problem is solved in 3 ways.

1) Have an initiative system plus ability to make the AI unpredictable (That means multiple AI cards for each enemy...which may be difficult to design). It truly is the superior way.

2 & 3) Regardless of #1, AI should control an enemy on its turn and AI should also be specific to the mission being played. An Enemy can't have the same AI from mission to mission. It will make the missions bland and feel like you've done it before. Even if it is as simple as changing priority targeting, etc. AI should change based on the different mission goals. That is what keeps the game fresh and exciting.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Mark Bradfield
United Kingdom
Pennsylvania
flag msg tools
Also posted on Alien behaviours
Thanks for that - I just read the Q&A today just before I posted.

It's the behaviours I am interested in.
If we could get a handful of the alien cards that gave their behaviours (reveal- offence - defence - swarm/reinforce - search/destroy ) - and a couple of tiles (which we have) we could come up with a workable AI based upon a flow chart & cards (that would add a random element) that would give the AI some teeth - even on the standard tiles.

The behaviours could be modified by the mission cards for the AI.

Like behaviours.
1. Reveal behaviour - instinctual attack/defence/run for cover/run to nearest ally.
2. Attack nearest enemy If melee is strongest attack then melee.
If ranged attack is strongest then ranged attack
If last card was swarm card - then move towardss and attack nearest enemy.
3. Defence - seek cover - move to nearest cover.
If last card was a swarm card then move grunts between highest value asset and heroes - hive mind instructs grunts to protect higher valued friendly assets.
Attack nearest Hero if possible
4.Swarm/reinforce - Move towards nearest friendly. Pick highest value friendly if there are multiple friendlies within range.
Once aliens have successfully swarmed they are considered to be a single unit as far as AI is concerned (ie they only get a single AI card from there on.
Attack nearest hero if possible AFTER Move has been made.
5.Search and Destroy - move towards map objectives - destroy before heroes can rescue/use.
If heroes achieve nearest objective then draw another AI card.
6.Divide and Conquer - a Squad of aliens split up - each figure then draws an AI card and follows those orders.


Once a Card's instructions have been completed for any particular alien.
Draw another card


Just shooting from the hip...
This would necessitate a ranking number on the Alien cards.
This would maybe take out the need for an actual flowchart.
Between alien behaviours and AI cards this might actually give the heroes a reasonable run for their money.

 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Mark Bradfield
United Kingdom
Pennsylvania
flag msg tools
Sorry just remembered a note from someone, somewhere about the aliens not using scrap or found items.

Change search and destroy to picking a Heroes and then concentrating firepower attacks on them?
Pick by rolling D6 and picking the D6th hero as target?
or roll the game dice and figuring out a way to count using them.

just a thought
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Thrawn 007
msg tools
The biggest revelation for me for Solo/Coop is that balance is no longer as much of a factor when designing this style of play. In head to head play, there is a lot of constraints that must be followed so that both the alien and hero players have a chance to win. Making one side or the other too strong is going to cause issues.

In coop mode, everyone is playing the hero side. So making the "puzzle" of the alien side more complex, allowing it "cheats" etc is much more acceptable.

I think this would mostly factor into the strategic level of the game, but it can also play a part in the tactical play, allowing aliens to do extra things to make up for not having a cohesive strategy.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Front Page | Welcome | Contact | Privacy Policy | Terms of Service | Advertise | Support BGG | Feeds RSS
Geekdo, BoardGameGeek, the Geekdo logo, and the BoardGameGeek logo are trademarks of BoardGameGeek, LLC.