It doesn’t look visually enticing, but I am pretty sure its darn good. No, I have not played the game yet.
Now, I wasn’t expecting to write a post on a game, any time soon. Not even on DotA. But this is just too good a game for me to resist. I have been quite intrigued by the concept of Evolutionary Computation. One lazy Sunday, I started looking for more stuff on Evolutionary Computation. I found a lot. Particularly, a post-doctoral researcher’s blog. Julian Togelius. He’s published a paper on solving Sudoku using Genetic Algorithms. Serious stuff that! One of his other papers have been judged as one of the top ten AI game innovations of the year. This dude was one of the organisers of a Simulated Car Racing Championship atWCCI 2008. It seems there are not a lot of games available that actually use the latest Learning Techniques.
It was very funny in the beginning when I looked at the games’ graphics( WCCI ). Not a ‘Need for Speed’ then. On further reading, I found more on the rules of the game. From a programmer’s stand point, I thought, not very easy to program a controller. All the various cases. How could it adapt?? Perhaps learn?? Neural Networks then?? Neural Networks are to put it very very simple terms, like a brain. ( Please note that this is a gross over-estimation, it resembles the brain only in that it can learn new stuff. It is not even close to the functioning of a brain ). Perhaps neural networks would have been the answer if the same question had been put forth a few years ago. Today, ANNs alone will not suffice!!
Genetic Algorithms are now used to train Neural Networks. OK. Imagine you are given the job of teaching something to someone. If you couple GA with ANN, it’d something like, you just have to tell the person, how well he has learnt the subject, and he’ll keep improving all by himself. ( Of course, this is another gross over-estimation. If this is very non-technical please refer to one of the links given in the side )
It just got more interesting after that. I learnt that one of the Car controllers was evolved using rtNEAT( real time NeuroEvolving Augmenting Technologies ) developed by Kenneth O Stanley at University of Texas at Austin. NEAT not only evolves the best neural network of a given structure, it also evolves the best structure for the purpose. Fantastic stuff. As I kept reading, link after link. I get here. The page gives a detail overview of how NERO is played and how it is being developed. Please do read it. Its an interesting read. You’ll start wondering what kind of games will be the next big trend.
I have added on my links lists some of the more interesting blogs I have found during my research. Anyone interested in AI, please do have a look at these pages. I definitely plan on writing more posts on my own discoveries on the Web about AI, more particularly, Evolutionary Computing. And next time, I hope I try to convey more technically what I actually wish to convey. Cheers.