Mellow Yellow

[Linguica]

Captain Mellow has been busy lately! On his Bot Page he has posted a "retrospective" on all the Doom bots which have been in any stage of production: the Nexus Bot, the Gravity Bot, the Doombot, the Cajun Bot, the Deathbot, the Thrust Bot, and the Skulltag bots. Bet you didn't know there were quite so many, eh? Also, Mellow Land Productions talks a little about some mods he's been making, not to mention a Doom sidescroller he's working on, of all things.

[NightFang]

Hey Captin', my name is NightFang, not DarkFang. Get with the times! :)

FIRST POST!

[Guest David_A]

I remember Grave... Didn't he quit Doom to go work on Golgotha (Crack Dot Com's unfinished game that they released for free when they went belly up)? That game wasn't exactly the shining moment of the open source community (ie nothing ever happened to it).

[aurikan]

as long as we're pointing out naming errors, i'm "aurikan" not "Aurican" ...

oh yeah and i'm disappointed that you STILL don't have a link up to my white paper on bot theory. good stuff there.

--> http://frad.org/oldstuff/bot/

[Guest Anonymous User]

Aurikan,

You make some interesting points but you make an assumption that is incorrect. You define intelligence in terms of actions (roaming, combat, and appraisal); that is, make a bot do something and it has intelligence.

A truly "intelligent" bot (which strictly speaking is not possible to do with todays technology; we can only simulate intelligence) actually does something because of its AI; the actions flow from a higher order decision making process if the bot.

An example will make this clear: a bot following a path node from A to B is not intelligent. It is simply folowing a set of points. If on the other hand a bot follows point A rather than point B, because B represents a higher risk than A, then the bot has displayed a semblance of intelligence.

To sum up: actions are the result of intelligence, not the components of intelligence.

[aurikan]

i'm not here do get into a flame war about what is or isn't intelligence in a bot

however consider that the bot does have some sort of neural network or weighted choice-making algorithm that leads him to choose a point C over point B, when the bot is at point A. How is he to get to that point? That is the point that the white paper intends to address - not the rather nebulous (and, without efficient and effective pathfinding, entirely irrelevant) subject of decision making.

[Guest Anonymous User]

Nightfang: I was wondering if that was you. Why'd you go & confuse a poor guy like that? Anyway, I would still stand by the fact that at the time if the Thrust bots, your nick _was_ Darkfang. So, let's see it as a historical reference. ;-)

Aurikan: My apologies for the mangling of your moniker. However, I think it would be nice to see you put all that theory to the test. That way, you don't come off as pedantic. Why don't you make a functional bot that is at least on par with Fly's or Yoghurt's work? As it is, I cannot take seriously your bot theory unless you have gone as far as Yoghurt or Fly. Walk the talk buddy. Fly's & Yoghurt's work has spoken volumes that your platitudes & incomplete NBot cannot yet approach. C'mon man, enough posturing. Do it.
-Capt.
btw Abuse (the game) is the bomb.

[aurikan]

Sorry, but i'm not going to spend my time implementing my theories when i have a real job and university to worry about.

Any serious and thorough evaluation of my application and extension of A* Algorithm concepts to a continuous plane would reveal that it isn't just a lot of hot air. But don't expect me to work to prove it to you. However, the mere fact that you run a bot resource page should be enough to merit a mention of 'possibly good' information.

[Guest Anonymous User]

Awwww c'mon, I double-dog dare you to make another bot. You have the time, you just need me to help with your time-management.

Here's your new itenerary:

1. Close Xircon for one week
2. stop updating web pages for one week
3. pour all the energy spent on the two above items into making bot theories a reality

And since I'm not as smart as you (heh I took dummy algebra in HS w/ all the other burnouts), I cannot abstract those theories you expounded. I need a concrete bot. Hey, now there's a good name.

-Capt.

[aurikan]

Little do you know that I am only really chatting in IRC 1) while the code changes I make at work are compiling and 2) when i'm taking a short break in order to keep my (in)sanity.

As for my webpage, if you look closely you'll notice that updates have all but stopped over the past 2 weeks. Even on a weekend when you think i'd have spare time i'm ditching everything to go to Toronto to enjoy my birthday and discuss marketing and future company plans with the co-owners of my dot-com business. If i had the energy to do anything else believe me, there are a lot of other projects in the line before doom bots. (Gruftistats-perl comes to mind ... as it is only 1 showstopper from beta release).

The intention of publishing the white paper in the first place was to provide any future bot programmers with some sound algorithms so that they wouldn't have to reinvent the wheel -- without forcing me to spend the time actually implementing it!

[Wildman]

Aurikan,

I am sorry you mistake open, honest discussion as flaming. It makes me wonder how seriously you take you're own white paper.

I have personally am quite familiar with AI having built more than one decision making and rule base system.

I am familiar with fuzzy logic, genetic algorithms, classifier systems and neural networks. I have used LISP, Prolog, C and Visual Basic (believe it or not) to implement AI systems.

I just finished a project that analsysis applications based on rules. The rules are several megs worth of data. My system can produce an evaluation on the input data in 6 seconds of real time.

The truth is that the mechanics of the game (path following for example) are trivial. It is not hard to make an automaton follow a path. I can give you a simple rule that will solve virually all closed mazes: always take the right hand branch. Implement it as a recursive routine and your done.

If you look at AI from a systems approach you can break it down into these parts:
Analysis; Decision; Execution; Evaluation; Learning.

1) Analysis: analyse both internal data (a node path) and external data (I am at node a).

2) Take the analysed data and apply a rule: The rule output will be the decision path. (Which way should I go based on the data?)

3) Execute the plan of action. (Go right because the path is shorter than left.)

4) Evaluate the results. Were the results as expected or not? (I cut off the player and attacked him doing X damage.)

5) Learning. Take the good and bad feedback and modify exisitng rules or create new ones. (If the player takes the left hand path, take the right hand path.)

Notice that this outiline emcompases both the internal intelligence and the outward actions. I think it is far from trivial. :)

[aurikan]

Wildman, I wonder what purpose your message serves. At what point did this become a flame war? It's more along the lines of Captain Mellow trying to egg me into implementing some of my ideas.

However, statements such as "I am sorry you mistake open, honest discussion as flaming." suggest to me that you aren't approaching this open, honest discussion with an open mind.

"It makes me wonder how seriously you take you're own white paper." -- I'm not sure where this remark came from. I wrote the linked article a long time ago, when I was doing research mostly on the navigation algorithm. Whether my proposals work efficiently and effectively is largely untested, though personally I believe them to be correct.

Following that, your post contains a large amount of fairly irrelevant information to the discussion. I don't think you realize the point of the white paper - to propose an algorithm for finding a path to follow once the decision has been made to go to point b from point a. You follow with an accurate if irrelevant analysis that looks at the idea of the systems involved in a bot.

Perhaps a better thing to do with this information would be to flesh it out with details and examples and write it up as a white paper, rather than post it here. If you want to discuss why or why not path FINDING is trivial, or the algorithm I have proposed, I would be happy to via email.

On an interesting sidenote, I implemented the proposed navigation algorithm on my TI-89 and gave a talk on the subject at a mathematics convention a few years ago.