Update to my Vindinium Client

Mon 20 October 2014 Free Time AI / Java / Vindinium

Features

There are two different interfaces for writing a bot:  SimpleBot and AdvancedBot.  The SimpleBot interface is just a Java representation of the GameState as defined by Vindinium.  As such, the map is represented as a string, forcing each bot to do its own parsing, etc.  AdvancedBot takes this further with an AdvancedGameState, which provides a board graph and provides a hash map for indexing the players, mines, and pubs by position.

Each bot package also comes with a callable to run the game loop.  The Callable will handle parsing the game state, invoking the bot, and communication.  Once a bot is written, it just needs to be instantiated and passed to the appropriate bot runner.

For ease of use, a Main class is provided that can run a bot on the class path.

Example Bots

Judging by my bot's score, I'm no bot expert.  Still, I included some examples.

SimpleMurderBot does one thing:  it shanks.  It does a simple Dijkstra search on the board and looks for the closest bot to stab.  If it gets low on health, it'll look for the nearest tavern.  It does not discriminate or take mines.  It literally just goes after the nearest bot.

AdvancedMurderBot is a little more interesting.  It also uses Dijkstra to do its search, but its decisioning is more advanced.  After reading about Behavior Trees, I decided to try my hand at implementing them.  Because this game isn't real-time, the implementation was much simpler than some of the examples I came across.  I have a Decision class that, given some state, will return some move.  The state and move are generics so that other bot designers might use different game state types.  As an example of why someone might want to do this, I created an inner class within AdvancedMurderBot that contains the AdvancedGameState and a Dijkstra search.  This enriched game state is what I pass to each of my decisioners.

In my implementation, I wrote several Decision nodes, each of which might return a move itself or decide to delegate to some other decision node.  The nodes could be assembled any which way.  In my case, I chose to assemble them as a DAG in order to make them easier to reason about.

Both bots also contain some pretty bad flaws.  The biggest of these is definitely the routing.  SimpleMurderBot will simply try to walk through other bots and doesn't consider them obstacles.  AdvancedMurderBot will avoid other bots that aren't targets, but it won't walk far enough around them to avoid a fight.  Rather, it will just walk next to them, provoking a strike.  Another interesting flaw is the lack of loop detection.  Both bots are vulnerable to getting into infinite loops at the pub with other bots.

Go Forth and Extend

I mentioned in my README that I think the AdvancedGameState still leaves something to be desired.  I think that my routing is pretty crummy, too, so I'll be working on those things for my next release.  If you spot a flaw, something that you think just sucks, or an improvement, feel free to send me a pull request.  I'd be happy to have your improvements.