New Competition from the guys of aigamedev.com

[delt0r]

Thought I would draw your attention to another competition at least partly inspired by the AI Challenge.

http://aisandbox.com/

Its a real time squad capture the flag type game and currently supports python.

I have seen one of the earlier versions of starter bots games in the player and it looks very interesting.

Full disclosure, I don't work for aigamedev.com, but am friends with a few of the guys working on this.

[moduspwned]

A shame it is windows only. The greatest part of aichallenge.org was you could write in any language and the environment worked anywhere.

I don't have access to windows, oh well. I will keep looking for other fun contests.

[lorill]

Their game looks nice, but yeah, mono language, mono os for now, and their starter kit wasn't tested recently... Their competition script fails, and their empty bot uses classes that seem to have disappeared or changed since it was written.

Small things to fix, but it doesn't give the professional style their website tries to emphasize. Oh, and there is absolutely no information about the available processing time...

[pdunstan]

Hi guys,

I'm pleased that you have seen it and downloaded it. If you'd let me, I'd like to assuage some of your fears.

We plan to release Linux support and the first of our language starter kits (C++) on Wednesday. We will be officially releasing the documentation for the network communication protocol between the commanders and the game early next week, and want to work with the community to support additional languages. This process has already begun but we are unable to announce anything at this time.

The problems with the competition.bat script and the empty commander class will be released in an update on Monday and a workaround for the competition.bat file is documented in the known issues section of the site (http://aisandbox.com/issues/).

There are no limitations on computation time, with the proviso that each commander is limited to one thread. The game runs in real-time; the game does not block while waiting for orders to be received. If an entry wants to spend three seconds analysing the current game state then that is there choice, though the game will have continued without them.

We know we are a long way behind what the AI Challenge has built up in functionality over the past three years but we are certain that we can host an interesting and fun competition.

Thanks for your input. I hope to see some of you in the competition.

Philip Dunstan
AiGameDev.com Team

[richardk]

Hi everyone,
I am sorry to see that our competition was dismissed so quickly here. We opted for a release-early-release often philosophy and hoped that we get a lot of people onboard to help us shape a great competition. We are a small team and thought this is the way to go - we might have been wrong.

Nevertheless we released a new version of the sandbox and the capture the flag SDK yesterday including a Linux64 version and a C++ starter kit and a description of the JSON network protocol. A GO starter kit was released by a community member and a Java starter kit is on the way.
http://aisandbox.com
So I hope we might be able to get some of your attention (back) - we are working hard to improve the ctf competition.

Thanks for your time,
Richard
AiGameDev.com Team

[smiley1983]

I would not be discouraged by the lack of response here: these forums move very slowly when there is no contest happening.

If dismissive comments have been made in IRC, it simply means that people are wary of putting effort into writing code for a platform which is still under development.

From my perspective, there are three key things which would make the system much more attractive to use. These reflect my personal priorities, and are as follows:

1) Separating replays from game simulations. As it stands on my system, I can only run the simulations at proper speed with no display ("headless" mode). Clearly I need to see the replays of these games, since raw win/loss data is of little value by itself. I understand that you are already planning to have replays, so this point is just highlighting one thing stopping me from starting development immediately.

2) More game specification data. Movement speed, stopping delays, turning delays, defense vs attack vs charge delays - and since this data may be in flux, it should be passed to the bots as setup variables. I know this has been discussed on IRC already, and you have reasons for not wanting to provide this. Here is how I see it:

If you don't provide this data, or provide it in documentation only, the result is that values get hard-coded into the bots (in one way or another) and they become brittle, responding poorly to changes in the setup, unless the developer comes back and re-anneals the hard-coded values for the new situation. I'm disinclined to work on code when that's the future I see for it.

I understand that certain general principles and relationships are being maintained, and that this data is available. The problem is that the underlying game is a simulation, and the bots which perform best will often be the ones which judge the edge cases most accurately - this means you benefit from having the real data, and if it isn't provided, it creates an incentive to extract as much as you can from the engine through some sort of annealing process. This is processor-intensive and not fun at all.

3) Some assurance that the source code will be released at some point. Again just my personal priority, but I'm wary of putting effort into closed source endeavours, because their survival and future path is entirely dependant on the decisions of whoever owns the source. EDIT - this is not a promise to put work in if it does become open source, just stating that when the source is closed, it looks less appealing to work on.

I also imagine that placing a link to the forums in the header of the aisandbox website would help people find them, and see that there are people developing bots. Talk of future contest ideas would probably also get some attention.

I think the project is a great idea and shall be following its development. Interest will pick up over time; there were people discovering Ants after it was over, and disappointed to find that it was too late. Word spreads slowly in these parts, it seems.