I take that back. Kim's stuff is mentioned at the geek site, but without a link. Kim?
Yes, the project doesn't really have a home page. Rob has put up some info about it into the geek site, although some of it might be currently a bit outdated. I can try to cover some status of the project here. You'll also probably realize that this is not such "small" GPS project that greggerm
was initially asking
To begin with -- the project has gone through a lot of experiments. It started off with a Palm Vx interface using IrDA to transfer raster map data from empeg to the handheld. Recently the interface was upgraded to use Compaq iPaq as a handheld which naturally provided full color map browsing and other neat features which were impossible to achieve with Palm due to the slow CPU (iPaq's SA1110 @ 206MHz CPU being 10x more powerful than Palm's DragonBall @ ~20MHz).
However, after a lot of experimenting and trying to find the most practical implementation, the handheld interfaces has proven to be more a cool/geek factor thing than a simple and effective way of interacting with empeg. Therefore the current direction would seem to be to embed all that interaction stuff directly into empeg and using the existing front panel buttons and remote as an interface.
On the navigation side, most of the stuff is based on vector maps which enables real route planning and navigation based on actual street names and locations. This part has been working for a while now - although - trying to find the right patterns for proper navigation controlled by speech is a challenge, taking the input data restrictions into account.
I've also spent a great amout of time to make the navigation module to run in background with the player fluently. In practice, this means kernel modifications to share audio, display and input parts with the player. This is quite essential bit, because you still want to mainly listen music in car and presume the navigation to be a background task. As along other things, this has been a challenge too, but has proven to work better than I thought it would.
While all this sounds nice, it's currently not easily portable to other environments. Even the software installation procedure (including custom kernels, certain requirements on permanently RW mounted scratch partitions, etc) makes it rather nasty to play with. Although, the biggest bottleneck probably still being the vectorized street level maps. As there hasn't been really a global standard for such data, it mainly works with a one dataset unless specific conversion is provided. I'm also aware of SDAL format, which is an attempt to uniform the "standards", but it'll still take time for bigger companies to adopt it internationally. Also, the availability of proper vector maps greatly depends on geographical location, making the situation quite a bit different to US and Europe, for example.
However, it's an interesting project that I'll keep working on, but unfortunately it won't be something that I could give out publicly at this point, partially due to the reasons above.