|
 |
TBM vs Others |
|
Ill try and keep this nice and simple, so it can provide an interesting read to those who dont want to get too deep (and also so I dont give any secrets away :P)
for those who dont know much about Artificial Intelligence (AI) ill give you a quick run down of whats what.
Fuzzy Logic: Most people who're here will know that binary is a set of 0's and 1's, fuzzy logic is basically a fancy name for grouping numbers into sets.
Expert system: Still researching this one, basically, you give it a set of correct questions and answers and it can then perform things like diagnosis.
Neural network: Looks like the best of the bunch, very powerfull, but somewhat limited in its speed and capabilities as far as I can tell. take a look here for more info
One of the things I purposefully did when developing the ideas for the TBM_AI system, was to avoid pollution from other AI systems, given its now pretty much proved itself under various test conditions, I've been testing other technologies and developements while I find the direction I want to go. (the two that seem the most promising are the raytrace - openRT and various audio systems such as openAL and VOIP)
So where does the TBM_AI fit into all this?
well, what you've seen so far is the toolset developement, I purposefully minimized exposure to whats truely going on in the core developement (if anything its been the TBM_AI ruleset thats been doing most of the public work on ltktbm), the main exposure you've had to the TBM_AI has been the pathfinder project and its optimizations (2 minates to build route info on a map was still with us not 18 months ago, and the capability 1.8 now has still doesn't really exist anywhere even now) and some small hidden tweeks to the AI which stops them behaving like complete fools.
To get you thinking a little more of where I plan to take this stuff, let me point you to two links:
game of life
this beauty is an expansion on part of the inspiration for TBM_AI
Artificial Neural Networks Technology -training
and I quote
| Quote: | | At the present time, unsupervised learning is not well understood. This adaption to the environment is the promise which would enable science fiction types of robots to continually learn on their own as they encounter new situations and new environments. Life is filled with situations where exact training sets do not exist. Some of these situations involve military action where new combat techniques and new weapons might be encountered |
And this my friends, is where TBM_AI fits in, beyond implimenting special information on the map for optimisation purposes, the TBM_AI system is completely unsupervised, I remember back in the day the week after the first TBM_AI system when I first saw a bot hide behind a tree in teamjungle while developing its tactics, after less than 8 generations - completely learnt behaviour (it shouldn't even of been able to know the tree existed) and I nearly fell off my chair.
Why are we not there yet? well mostly set backs. The original code base was Extremely buggy, crashed alot and was generally unusable for anything other than simple analysis, and no-one was interested.
So how far on am I? pathfinder, is basically done, and this is a big boost for the real-time processing required to make the AI SDK a reality. The ruleset is defined in my head, and I have about 30 pages of handwritten notes on how to impliment it in code, providing for maximum versatility, theres many applications it can handle, and as such needs to be written with these in mind. Most of the tools and principles have been well tested, but final versatility has yet to be realised. Version 1.9 should mark the end of this stage, and maybee the dawn of a new day for TBM in general. Especially with the parrallel developements such as q2a3d and the water surfaces. It also has shown promising results for the likes of project and financial management, which may prove to be its most usefull functions.
So what next? Ill not go into to much depth as to where I plan to take it, other than to say this. Yesterday I read an old (1995) publication called "Fuzzy Logic, Neural Networks and evolutionary computation" this made some pretty fantasical claims based on research at the time and where they expect to be using 2001 technology. Well the technology is here, above and beyond what they were looking for. But needless to say it didn't live up to expectations. With this in mind, The TBM_AI SDK should provide technology sets beyond my expectations, but untill its tested more throughally all I can say is hope. However, a nice discovery recently is how well neural networks can provide a 'plugin' to the current TBM capabilities - they fill in some irratating holes with regards to recognition and fuzzy logic. these WILL be a part of the SDK, so the very least you can expect is unrivelled pathfinding capabilities, with some well proven learning systems.
Why donate? Well, your not just saying "thank you" for the time Ive already spent to bring you what I would hope is some great entertainment, but also ensuring I have the time and resources to continue todo so. Let me get this straight, I may be getting on a bit, but I am a full time student, you may have seen me less and less recently (mostly due to part time work to fund my degree), but what you have seen I hope you like, the more donations I recieve, the more time I can spend evolving what we have already, and believe me when I say you aint seen nothin yet. Lastly, To become a beta tester from now on your probably going to have to donate. This means, if you want access to the water surface tech demo stuff, its source, and maybee even the ltktbm beta your going to have to donate.
Comment on this article (1 Comment) |
| | |
|
|
Word of the week |
| |
|
