Among my many dumb ideas is a take-off on the Forza Horizon concept. One of the differences, however, would be using real roads, scanned in 100% as-is, rather than a condensed and/or modified version of an area. The problem is that some areas are pretty dang big. Examples of what I'm talking about: Kanto, the alps of northern Italy (i.e. Stelvio and the surrounding area), south-central Alaska (yup, I'd do that). So, how much land area could a modern racing game depict at a high level of detail before overloading the computer/console it runs on, or running out of disc space?
Scanning in and committing real world terrain as mesh data would result in insane amounts of data, meaning it's highly unlikely that you would see scanned real world locations outside of something like track-based racing games, which recreate existing, limited area racing facilities with a high level of accuracy. Procedural generation, on the other hand, could effectively generate anything up to a full scale earth, or more. Take, for instance, the offroad racing game FUEL, which had, if I recall, a 14 thousand kilometer square environment, using procedural generation layered with some pre-built assets. There also already exist engines like Outerra, that can generate a full scale earth-sized environment, albeit without roads or other granular details and such at its current stage. In the current state of the technology available, it's not hard to imagine a procedural generation system that could, say, apply satellite mapping data of roadways and topographic information to a terrain generation engine, creating at least a rough approximation of a full scale earth for racing games. It would, of course, be an extremely complex technical endeavor, however it is not outside of the reach of modern technology in the field.
I'm not exactly aiming for a full-scale Earth. I'm thinking more about featuring the car culture of a specific geographic region. All of those areas have some problems that would be annoying though. For Kanto, the problem would be endless mazes of impossibly narrow alleys and clogged surface streets. For SC Alaska, there are very few truly fun roads; most of them are in deplorable condition, are residential, or don't lead anywhere interesting. Also, it would be very difficult to avoid terminating the map in a dead end once you got way out in the sticks. For northern Italy, the problem is, quite simply, finding a place to terminate the map... its problem is the exact opposite of Alaska's, in that you pretty much can't throw a rock without hitting a slice of asphalt heaven. It just goes on and on and on. Even the freeways there can be interesting - as you go further south toward the coast, you'll find a highway on which the northbound and southbound runs have completely different layouts, probably due to the terrain being moved around/through.
To correct a few things in this post, Outerra isn't exactly fully procedural. It uses scanned real-world data, collected from such sources as NASA's SRTM. What that lacks in fidelity (most data is 30-meter resolution) Outerra fills in with procedural detail down to the milimeter. Which does make it extremely granular, but, yes roads, rivers, and buildings are indeed a no-show by default. Though those can be constructed in Outerra's built-in editor. FUEL also used pre-designed terrain for most of its larger features. There has been some discussion on Outerra's forums about importing road data from Open Street map, and tests using selectively higher-resolution (10-meter) data for terrain generation. There have also been some rumors to suggest that Cameni and Tdev were in talks about perhaps combining their technologies, but nothing ever came of it. There is another experimental engine called Unigine that could simulate very large areas of land (Roughly the land area of Florida) without going so far as trying to simulate the entire planet. From what you describe, this is perfect for your concept. Your other complaints echo the difference between games and the real world. What's inconvenient and dangerous to real people is also interesting to if not downright necessary for the attention of players. I guess that means that us level designers are here to stay.
