troyv wrote:And if the surface of the road is poor (not talking about 4x4 routes), that means that the speeds on those roads will be lower, and thus, Waze will learn to avoid those routes.
One problem with that logic: poorly-maintained gravel roads inherently have fewer drivers using them. That reduces the odds that any of those drivers have Waze running while they drive on the poorly-maintained gravel roads. That means it's entirely likely that Waze won't have any baseline speed data for the segment(s), and therefore won't know to avoid them.
I agree with your argument and here is why.
So my original thinking was that if we were to use a network topology as an analogy, the router (Waze) would just send packets (users) down a certain segment (road) and see what the results were. If the packets (users) made it down a certain segment (road) OK, then the router would send more. That's great in the network world, but if Waze were to guide a user on to a rocky hiking trail in the forest using their Ford Taurus, that would be a bad thing. Basically, we shouldn't treat users as packets in a network.
Now I am thinking out loud a little bit..
What if we took advantage of users and had a "get me out of here" button on the client. If that button is hit, Waze would attempt to guide the user to the closest well driven road as quickly as possible. Not only would it do that but it would take a note as to what road the user was on. When Waze was tempted to route another user that same way, it would see that a previous user wanted off of that road. Waze then would have learned a little bit to avoid that road segment when routing for future users.
That's my $.03 (Have to account for inflation)