It would seem to me that the dynamic nature of Ruby would make something like this pretty cool.

Thanks.

Are you interested in writing it?

If helping build a training tool like this helps me learn then I'm all for it.

What would we need just to start? Who's interested?

Kyle Heon
kh...@comcast.net

Cheers-
-Ezra

On Nov 9, 2005, at 5:02 PM, Kyle Heon wrote:

If you just want a port of robocode, I'd say a good approach would be to
take the existing robocode source code and replace existing classes with
Jruby. Once you've got the majority in Jruby, refactor that around to
make it more rubyish.

I'd imagine straight porting of the majority of the code would be a
fairly simple matter - it would only be once you started porting the GUI
that you'd be into hard stuff.

I'd also imagine that instead of running the code in a sandbox, you'd
instead expose something over dRB/some other sort of connection for
clients to connect to and fight it out. This would allow you to write an
adapter than would allow an existing robocode robot to enter the fray
(which would be a great kick-start, and also a great way to test).
Alternatively, you could just eval client code with appropriate $SAFE
levels (which would probably restrict cheating more). Or you could do
both.

Otherwise, you'd need to hash out a plan for a new environment for AI's
to play in - Maybe some sort of random maze to fight in? - Maybe a
capture the flag game?

Another fun AI thing to do is to write AI's for simple board/card games
(ie Ruby Quiz #51), although watching computer players play isn't
usually that much fun, it's much easier to insert a human into the game
that is turn-based, and it's usually not that hard to play in a
text-based environ. Poke around boardgamegeek and find something you
like and build a two player version with a nice interface and then send
it out for people to write AI's for. If you go this path, write your
game rules in a server program and have all the capturing of input and
displaying of things in client programs - this will normally you a
decent interface to build an AI on. Something like Carcassone has
extremely simple rules, but can provide a real challenge to win.

Even figuring out the winning rules to solo games such as Solitaire or
Sliding Puzzles can be a challenge, but more fun is had when there
unknown is in the form of another player (rather than just figuring out
statistics, which solves most solo games).

First, another existing project to check out is realtimebattle. I think
you could write something in ruby for that.

Jacob and I wrote ruby agents for an AI class using TCP sockets to the
capture the flag game server which has probably been retired since they
now do BZFlag agents instead of the shoddy CTF server we used then. But
the premise was simple enough: you talk to the server over a tcp
socket. You get precepts and send actions. I think this is far and away
the best approach. Yes, then someone can use any language they like,
and I think this is a good thing. It certainly helps not to get in the
way of its growing popularity.

The questions then go to what precepts and what actions. It is my
opinion that most people will be coming into a game like this wanting
to do something interesting as early as possible. Otherwise it will not
be fun and there will be a lot of half-written bots and few real ones.
So simplicity is key. If it were a serious AI tool you'd probably want
to make the precepts and so on configurable, but I say let it evolve
there if driven to.

A rough description of the game world I decided would probably be the
most fun follows.

The world is a 3D sphere in space. No gravity sources of consequence
are nearby. The agents are spherical (cow-shaped) pods with yaw, roll,
and pitch thrusters, and a front-mounted laser. When you shoot another
pod, the pod is frozen (can't fire thrusters or lasers) until a
teammate unfreezes it in the same manner. Flags are magnetic and so
when you get close enough, if your magnetic field is on, you have the
flag. You win by bringing the enemy's flag back to your base (a
predefined spherical region). Two teams. Each pod has an
omnidirectional limited range radar (does such a thing exist in the
real world? I have no clue), which is the only means for observing
objects. For simplicity (maybe this would be taken out for expert mode)
you are told your precise position when asked. If you try to go out of
bounds, you just don't. (e.g. you keep your momentum in other
directions, but just stop going outwards).

The server wouldn't be too hard, the hard part would be the vector
stuff (e.g. did that laser hit anyone?) and being time-independent from
the network stuff.

The clients would be easy, but since you're dealing with 3D and limited
information it is still an interesting AI problem.

The real hard part (from my perspective, probably not from someone
else's) is the graphical part that lets us humans watch and/or
participate.

However, I'm very interested in a quiz to right the robots if someone  
gets an engine together...