Recommend
14 
 Thumb up
 Hide
3 Posts

Adrenaline» Forums » Variants

Subject: Brainstorming Bot AI for solo, 2, 3, or 4 players rss

Your Tags: Add tags
Popular Tags: [View All]
Mike Cook
United States
Spanaway
Washington
flag msg tools
Avatar
mbmbmbmbmb
So from the first time I heard about Adrenaline I have been slightly obsessed with it. A unique theme and combination of mechanics by one of my favorite publishers! I don't typically like any kind of area control, but this has elements of it in a package that seems like a blast (most area control games are far too dry for me). That means it not only seems like a great product, but a game that will occupy unclaimed territory within my collection.

The first several videos/interviews I saw hinted that, though it is a 3-5 player game, a 2 player variant or rule set was being tested. Since I play a good 75% or more of my games 2 player with my wife this a fairly important selling point. I think the only game in my collection that requires more than 2 players is No Thanks!. So when it was announced that CGE was not content putting out a simply passable version of a 2 player variant/rule set I understood, since CGE doesn't just make good games they make great polished games, but I knew that I love the sound of the game enough that I wanted to make it work with 2.

This thread is to jump start that endeavor. I know the game isn't even out yet, but the rules are posted and there are instructional videos. I won't be able to get the game until Christmas, likely, but there is enough out there that I wanted share my thoughts and to get the ball rolling so that this thing can hopefully exist by the time my wife and I get our hands on it! So lets do some brainstorming.

My first though is that the AI will be driven by a decision tree (duh), and my inspiration for that is BGG user RedJak7's Automated Imperial Variant for Star Wars: Imperial Assault. https://www.boardgamegeek.com/thread/1464659/redjaks-automat...
In this variant RedJak7 made AI cards with a prioritized decision tree for the activation of each imperial figure (there was plenty more to that excellent variant since it is a very different game, but this is the applicable element). When it was the given figure's turn you would simply start at the top of the card and perform the first two pertinent actions.

The one complaint I have with that system is that if the decision tree is known, and static, it is pretty predictable. So in my opinion creating one AI card, with one decision tree is not the best solution. An argument could be made that you could create different cards with differing decision trees to sort of act as different personalities or difficulties. To me that would be interesting and possibly viable, but still a little bit static, predictable, and dry in the long run. It also would require a fairly complex decision tree to account for all eventualities. That means more work to develop and balance, as well as more of a chore to step through in the actual game. It should be quick. So I think that there could be a small set of cards that basically prioritize specific actions/tactics. The final change I think from the Imperial Assault variant that inspires me is that you don't just start at the top of the tree for the first action, and continue down the list after the first action. Here each bot-turn you randomly choose one card and step down its simple decision tree for *each* action. So if the first action skipped decision 1 and did decision 2, you start again at decision 1 and proceed down the list again when determining decision 2.

These priority actions/tactics cards could look something like this one that focuses on reloading and being ready (off the cuff, rough generalization here) -

"Prepared for Anything" (or some flavorful tactic name)
If the player has any unloaded weapons get ammo necessary to reload and preserve this ammo for the turn
If the player has less than 3 weapons, a weapons station within range, and the ammo to do so, move and grab a new weapon
Priority - make sure player has at least one single-target and one multi-target weapon
If you have a loaded weapon, target in range (using adrenaline ability if available), and the ammo to reload, Shoot someone
Priority - First blood
Kill shot - overkill only if already possessing 3 marks from that player
Apply maximum damage (including marks) - use applicable abilities
Least damaged target - try to take the lead on one or more player
Move and Grab ammo
Priority - Ammo for desired weapons
Ammo for abilities on loaded weapons
ammo to reload loaded weapons


another could be this one that focuses on damage -
"Bring the Pain"
If the player has a loaded weapon and target in range, shoot someone
Priority - Apply maximum damage including marks and available abilities
Kill shot - overkill if possible
First Blood
Least damaged target
Move and Grab ammo
Priority - Ammo for abilities on loaded weapons
Ammo for fired weapons
ammo to reload loaded weapons
If all weapons are fired and ammo is full, move and grab a weapon
Priority - at least one single-target, and one multi-target weapon
Highest damage dealing weapon


The backside of all priority/tactic cards can have the printed tactic for Final Frenzy, which would probably just be a priority list for who to shoot unless you are out of ammo.

Even though there are only three actions to choose from you can break them up based on some criteria. What would need to be done is that all the options or reasons for getting ammo, weapons, and movement would need to be listed and then compile the different permutations. That said I'd be fairly surprised if anyone could really think of more than 10 of these decision trees without getting redundant or changing nothing but the shooting priorities. I'd think you would only need to make and tune about 6 or so of these tactics to get a full range of options. In fact that brings me to my other idea.

You could make those 6-8 or so priority action/tactic cards and then make a few priority target cards for determining who to shoot. So on a bot's turn you randomly choose a tactic card, and if you come to a decision that the bot will shoot, then you randomly choose a priority target card and use that decision tree to decide who to shoot. Some priority target cards could prove to be more difficult and if you like you could even just assign one to the bot for the game to serve as the bot's difficulty. Though I think it makes it a little more predicable and would personally enjoy the variety if you turn over the priority target card when the bot is actually shooting.

TLDR
I think the best solution is to have 6-8 or so (however many make sense without being redundant) priority action/tactic cards with a focused decision tree that will randomly be chosen for a bot's turn. This will guide their tactical decisions for that particular turn. Reload, fire a lot, get resources, be balanced, defensive, positioning, whatever. Then whenever the decision tree determines that you need to shoot someone you turn over a priority target card and it will help determine who is gonna get shot. Or deal a priority target card to the bot at the beginning of the game and let it be the bot's personality/difficulty/strategy. Then just use a priority action/tactic cards to determine the particular turn's tactics. You can use as many bots as you want in a game so that you have your favorite number of players and can play solo with up to 4 bots, or with 4 players and 1 bot.

Different cards would need to be developed entirely to deal with domination or turret mode.

What needs to be done is to decide on the different priorities for getting ammo, for getting weapons, when the bot may want to move to a more ideal spot (more or less densely populated), and some combinations of actions that make tactical sense. Then we need to think of all the possible permutations of priority targets (kill shot, with or without overkill, first blood, most damage possible, least damaged player...more?)

I'm sure there is a decent amount of factors I am not identifying and plenty within what I have thought of that needs to be fleshed out completely, but I would love to hear any input or other recommendations! Whether you think this system is a good idea, or whether you think something else is. A facet I'm ignoring or has room to be fleshed out more, etc.
11 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Mike Cook
United States
Spanaway
Washington
flag msg tools
Avatar
mbmbmbmbmb
Reserved

First very rough draft of the AI cards has been submitted to the files section and hopefully will be ready shortly. I will add a link when it is available.

Please do playtest, and comment with your concerns or ideas!
3 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Derek Funkhouser
United States
Columbia
South Carolina
flag msg tools
mbmbmbmbmb
Any more work on this? I just bought this game but I'm mostly interested in creating bots for the game to play against.
3 
 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.