.comment-link {margin-left:.6em;}

Sometimes I Wish That It Would Rain Here

Wednesday, September 21, 2005

AI on board

a couple of months ago, I had a post about the video game industry's focus on graphics over things like a physics engine or AI. one of the possible reasons I cited (other than the fact that graphics show an immediate, visible improvement) is that the performance block in graphics generally comes in the form of pushing matrix operations and a few other specific floating point calculations. these are (relatively) easy to create specialized hardware to perform. it may not be as easy to create the same specialized hardware for a PPU (physics processor unit), but it seems plausible if not possible. however, could you do the same for AI, create an AI accelerator chip/board?

that is exactly what company AISeek is proposing. they haven't made any technical details available yet (beyond an architecture diagram), but it looks like they will focus on four specific tasks: path finding, crowd and multiagent movement, terrain analysis, and sensory simulation. I can see these as problems that could be abstracted and hardware created to perform these tasks. furthermore, they are general enough to apply to a AI in a large variety of games. I wonder if you could also incorporate decision-tree algorithms like mini-max or a* into such hardware.

I don't know if this is going to happen, but I think it's rather exciting and a step in (more or less) the right direction for video games. now we just need someone to write some more innovative games.

in the hardware vein, I think hannibal has an interesting view, that these add-on functionalities will end up all on the same die as the CPU. just had to give the ars shout out.


Post a Comment

Links to this post:

Create a Link

<< Home