Recommend
9 
 Thumb up
 Hide
15 Posts

GMT COIN Series» Forums » General

Subject: Gandhi Solitaire Bots Wrapping up! rss

Your Tags: Add tags
Popular Tags: [View All]
Scott Mansfield
United States
Bend
Oregon
flag msg tools
designer
badge
Avatar
Microbadge: Solitaire WargamerMicrobadge: Holland '44Microbadge: The U.S. Civil War - GrantMicrobadge: Root fanMicrobadge: D-Day at Peleliu fan
Bruce has just finished the base design work on the solitaire bots for Gandhi. He's redeveloped the COIN Bots to be quicker and more dynamic than the current flowchart design.

Here's a little overview of how they work:

It is a hybrid system that uses a small card deck (9 instruction cards plus 1 shuffle card) to select Non-Player Operation + Special Activity, and provide the instructions for the first spaces selected.

Additional spaces are selected following an instruction set that is similar to exisitng COIN bot flowcharts (minus the diamond-shaped decision boxes).

This one decks runs all NP Factions. The cards in the deck are double-sided: one side of each card (the compromise side) has the NP Faction build its forces and board position; the flip side (the hard line side) has the NP play more aggressively. Cards are flipped as a result of the current game state (e.g., "Flip if Restraint is 0 to 2"). Each card also has a trigger, also based on the current game state (e.g., "Any Available Protest markers?"). When a NP Faction chooses Ops + SA, card triggers are checked consecutively. The first card that triggers determines the Op + SA for the active Faction.

The system is designed to be quick and provide more variety than the current flowchart-based COIN bots. In addition, NP Factions are limited by Lim Ops but do not spend Resources (Raj and Rev Ops are limited by a die roll). The bot is also programmed to go after the player factions to provide a challenging opponent.

I can tell you from playing it, it's fantastic and could be a COIN game changer in that it eliminates the traditional bot flowchart stagnation. There's limited mental calculations when figuring out what the bot will do.

Cheers,
Scott

8 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Galatolol 1
Poland
Kraków / Paris
flag msg tools
Avatar
Microbadge: I love FranceMicrobadge: 18xx fanMicrobadge: Historical games fanMicrobadge: Economic Games fanMicrobadge: Heavy Games fan
Looks like a neat solution.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Bobby Factor
United States
San Leandro
California
flag msg tools
badge
I am Volko Ruhnke's medulla oblongata....
Avatar
mbmbmb
I've never minded the flowcharts much (in point of fact, I'm glad they're included as they provide more-than-adequate opponents), but this new system looks very cool. Can't wait for this one to get printed.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Colin Taylor
United States
California
flag msg tools
Avatar
mbmbmbmbmb
Hi Scott,

Thanks for the information. Obviously, we haven't seen the full details on the bots yet, so read my comments below with that in mind.

Personally, I think there are 2 potential areas of improvement for COIN bots in general. First is the concern that for many, the bots are too complex and cumbersome to run, especially in the 4P titles, where a player may have to run 3 NP factions. Second is the concern that the bots are too predictable, essentially following one strategy the whole game.

Having seen some of the early ideas in All Bridges Burning, and your post here, the designers of both titles seemed to have zeroed in on the second area, but I wonder if the first has been addressed in any way? (note - I could be way out of date on my understanding for ABB, so apologies if that has changed). However, IMO, in order for the second area to be a large problem, it requires players to play the game a fair amount. And to play a fair amount, it requires less complex (or at least easier to run) bots. I am concerned that there are lots of potential players who balk at playing COIN solo (excluding playing all sides) due to the complexity, and therefore the second issue is moot.

What are your thoughts on this? Are there ways to reduce bot complexity without losing bot competitiveness? How much larger would the pool of potential COIN customers be if that were addressed?

Thanks,

Colin
5 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Scott Mansfield
United States
Bend
Oregon
flag msg tools
designer
badge
Avatar
Microbadge: Solitaire WargamerMicrobadge: Holland '44Microbadge: The U.S. Civil War - GrantMicrobadge: Root fanMicrobadge: D-Day at Peleliu fan
The Bobby Factor wrote:
I've never minded the flowcharts much (in point of fact, I'm glad they're included as they provide more-than-adequate opponents), but this new system looks very cool. Can't wait for this one to get printed.
Totally agree, but what we consistently see is the bogged-down nature they can put a player in. I've run 1:3 (me:bots) Fire in the Lake games where I'm spending most of my time deciphering the bots decision trees and less time playing my strategy. Not to pick on FTL for it's one of my all time favorite games!

What Bruce wanted to do is limit the time spent on figuring out what the bot does without limiting its quality of play.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Scott Mansfield
United States
Bend
Oregon
flag msg tools
designer
badge
Avatar
Microbadge: Solitaire WargamerMicrobadge: Holland '44Microbadge: The U.S. Civil War - GrantMicrobadge: Root fanMicrobadge: D-Day at Peleliu fan
These are precisely our concerns as well and the impetus for designing a fresh take on the design. The complexity question is an interesting one and I whole heartily agree. I'm a COIN veteran and I rarely play with the bots for this and the cumbersome nature of using them. I think the salability of a COIN bot design that addresses this is huge.

I have not looked into ABB and so won't comment on their work but on our end Bruce has done a few things that separate Gandhi. The primary one is the ability to take in the board state at a glance and figure the bots move quickly. Here's what he wanted to avoid; choose all spaces where X could happen if Y pieces are placed. In essence, to avoid the need to play through an Op mentally to see if it fits the flowcharts diamond. And in fact, take the diamonds away all together. So make the Bots decision easy to see and rapid to implement.

We looked quite heavily at the Conflict of Heroes solitaire card deck as inspiration, but in a 4-player setting we could only use some inspiration from it.

And because Bruce designed it all around a 9 two-sided card deck, the predictability of the bots has been drastically reduced and it responds to the human players moves in any given game.

Hope this answers some of your questions!

Edit:
1. Bruce tried to keep each NP instruction set limited to ease overhead, especially with the March/Demonstrate/Sweep Ops that can be quite complicated
2. His personal feeling is that solo games are best with a 20%-30% win rate; the bots are designed to provide a challenging opponent, not to replicate intelligent play
3. The base game itself was designed to offer limited but meaningful choices to the players, which in turn limits the decision tree for the ‘bots


ColintheFlea wrote:
Hi Scott,

Thanks for the information. Obviously, we haven't seen the full details on the bots yet, so read my comments below with that in mind.

Personally, I think there are 2 potential areas of improvement for COIN bots in general. First is the concern that for many, the bots are too complex and cumbersome to run, especially in the 4P titles, where a player may have to run 3 NP factions. Second is the concern that the bots are too predictable, essentially following one strategy the whole game.

Having seen some of the early ideas in All Bridges Burning, and your post here, the designers of both titles seemed to have zeroed in on the second area, but I wonder if the first has been addressed in any way? (note - I could be way out of date on my understanding for ABB, so apologies if that has changed). However, IMO, in order for the second area to be a large problem, it requires players to play the game a fair amount. And to play a fair amount, it requires less complex (or at least easier to run) bots. I am concerned that there are lots of potential players who balk at playing COIN solo (excluding playing all sides) due to the complexity, and therefore the second issue is moot.

What are your thoughts on this? Are there ways to reduce bot complexity without losing bot competitiveness? How much larger would the pool of potential COIN customers be if that were addressed?

Thanks,

Colin
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Colin Taylor
United States
California
flag msg tools
Avatar
mbmbmbmbmb
Well, that sounds extremely encouraging!

Thanks,

Colin
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Niko
Canada
Calgary
Alberta
flag msg tools
[This space intentionally left blank]
badge
Microbadge: Casual wargamerMicrobadge: COIN fan Microbadge: Band of Brothers fanMicrobadge: Space Empires: 4X fanMicrobadge: Motörhead fan
srmansfield wrote:
2. His personal feeling is that solo games are best with a 20%-30% win rate; the bots are designed to provide a challenging opponent, not to replicate intelligent play
And that is were you are starting to lose me. One of the big pluses was always that the COIN bots provided a challenge while playing by the same rules a human opponent would be playing.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Oerjan Ariander
Sweden
HUDDINGE
flag msg tools
Avatar
mbmbmbmb
Ze_German_Guy wrote:
srmansfield wrote:
2. His personal feeling is that solo games are best with a 20%-30% win rate; the bots are designed to provide a challenging opponent, not to replicate intelligent play
And that is were you are starting to lose me. One of the big pluses was always that the COIN bots provided a challenge while playing by the same rules a human opponent would be playing.
Do these bots work in 2- or 3-player games, or are they intended for solo play only?

How well do the cards handle urgent crises and fleeting opportunities? To take examples from FitL, if the VC has gotten Underground Guerrillas into a bunch of Support spaces with ARVN cubes the COIN Factions need to Sweep ASAP since otherwise they stand to lose lots of VPs very soon, and if NVA ever find themselves with a large stack of Troops in a space with a US Base they'd better Attack immediately since otherwise the US will probably destroy the entire stack in their next activation. To increase the chances of catching situations like that, the flowcharts check for them early. The card triggers OTOH are checked in a random order due to the cards being shuffled, so if an urgently needed action also happens to be fairly rare I'd worry that the cards would miss it more often than not because some other trigger usually comes up first.

***

The base game also has an impact on the bot complexity, of course. The main reasons why the 1st edition FitL bots are so complex - they are the heaviest in the series so far - is that each bot needs to watch out not only for itself but also for its ally, and that most of the Ops and SAs in FitL has several different options to choose between. To complicate matters even further, both sides have chains of actions that need to be taken in a certain order to be effective (with the longest chain being the COIN side's preparations to enable Civic Actions and - for the ARVN - eventually Govern for Patronage).

In Gandhi OTOH the Factions don't have any official allies so each bot can be more focussed on its own goals, Bruce has spent a great deal of effort to keep each individual action clean (IIRC max 2 possible effects in each action), and I also get the impression that the action chains are generally shorter. Each of these features by itself helps to keep the bot instructions simpler and cleaner; with all of them combined, it almost looks as if Bruce had the solo system in mind from the outset

Regards,
Oerjan

5 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Jason Carr
United States
Colorado Springs
Colorado
flag msg tools
Avatar
mbmbmbmbmb
Oerjan wrote:
it almost looks as if Bruce had the solo system in mind from the outset
Little known game history here, but Gandhi was originally a (withdrawn) entry for the BGG Solo PnP contest: https://boardgamegeek.com/thread/1397940/wip-hind-swaraj-ind...
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Jason Moore
Germany
March
Baden-Württemberg
flag msg tools
badge
"Playin' solitaire 'til dawn, with a deck of fifty-one."
Avatar
mbmbmbmbmb
Can't wait for this. The solo refinement sounds really intriguing.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Scott Mansfield
United States
Bend
Oregon
flag msg tools
designer
badge
Avatar
Microbadge: Solitaire WargamerMicrobadge: Holland '44Microbadge: The U.S. Civil War - GrantMicrobadge: Root fanMicrobadge: D-Day at Peleliu fan
Great responses! Just a few things to keep in mind. Bruce designed the base game with the solo game in mind. The above, "the bots are designed to provide a challenging opponent, not to replicate intelligent play" seems to have triggered a few responses. Bots can exhibit complex behavior, but not intelligence, which is the ability to think creatively through a problem. AlphaGo's famous move against Lee Sedol was not creative, it was surprising, but systematic.

And these will be the first COIN bots to follow the same rules as the human player; e.g. they may be forced to take a Limited Operation. Part of their strategy is that they treat a win by any bot as a win by all bots, by targeting player Factions first. This should keep the difficulty of any single Faction winning somewhat consistent between games with 1, 2, or 3 NP Factions.

Cheers!
3 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Niko
Canada
Calgary
Alberta
flag msg tools
[This space intentionally left blank]
badge
Microbadge: Casual wargamerMicrobadge: COIN fan Microbadge: Band of Brothers fanMicrobadge: Space Empires: 4X fanMicrobadge: Motörhead fan
srmansfield wrote:
Great responses! Just a few things to keep in mind. Bruce designed the base game with the solo game in mind. The above, "the bots are designed to provide a challenging opponent, not to replicate intelligent play" seems to have triggered a few responses. Bots can exhibit complex behavior, but not intelligence, which is the ability to think creatively through a problem. AlphaGo's famous move against Lee Sedol was not creative, it was surprising, but systematic.

And these will be the first COIN bots to follow the same rules as the human player; e.g. they may be forced to take a Limited Operation. Part of their strategy is that they treat a win by any bot as a win by all bots, by targeting player Factions first. This should keep the difficulty of any single Faction winning somewhat consistent between games with 1, 2, or 3 NP Factions.

Cheers!
Ok, I guess I misinterpreted that comment. I thought you were saying that they would not be simulating intelligent play as in not be following the same rules as a human player rather than in the trivial sense that cardboard driven instructions will not be intelligent...

Good point about the LimOp restriction, that has bugged me a little bit but I've considered it a necessary evil to ensure competitiveness of the bots without having the instructions become too convoluted.

Don't go too far with having the bots prioritize the players though, my understanding is that that was the biggest complaint about the original FitL bots... Later games like Falling Sky still have some degree of prioritizing the player, but will stop any faction from winning first. IMO that is a better way as it more closely sticks with how a player would act.

I want to emphasize that all of this is simply my preference; I want to play a solo COIN game as close as possible to a multiplayer COIN game. Not a solo game that kinda sorta is the same since the bots follow different rules.
There are already plenty of solo designed games out there that I can bust out if I want to play against a bot that follows different rules than I do, e.g. the D-Day series. I can't think of a game like that on the same topic as Ghandi though, so there will certainly be a place for it if you chose to go that route too.
3 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Gabriel Conroy
msg tools
mbmbmbmbmb
I've never been bothered by the bots' complexity - in fact I'd be happy with more complex bots! The thing that makes them drag sometimes for me is deciding which/how many spaces to operate in based on some criterion of whether certain actions are possible etc. It's unavoidable to some extent but I also feel some of it coy d be streamlined or even randomised without too much impact on the the end result.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Anton Tolman
United States
Orem
Utah
flag msg tools
Avatar
mbmbmbmbmb
This sounds terrific! Any update on their status? Any chance you might be willing to work on similar systems for other COIN games? Like, Pendragon, for instance -- just an idea!
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls