I think the key is "probabilistic models" and "sensor fusion". Not more accurate GPS. Let's say you know a guy is traveling at about 60 mph on a four lane road

I think the key is "probabilistic models" and "sensor fusion". Not more accurate GPS. Let's say you know a guy is traveling at about 60 mph on a four lane road, two lanes in each direction. Then the phone's acceleration sensors tell you the guy just moved "left" by 10 feet. A little while later you decide he just moved "right" by 10 feet. He goes back and forth a few times. You also make the assumption that he's not driving on either shoulder. So, you can guess that after he moves left he's in the fast lane; after he moves back right he's in the slow lane. Phones are so sophisticated these day. Here's something from Apple I found after a quick search: An accelerometer measures changes in velocity along one axis. All iOS devices have a three-axis accelerometer, which delivers acceleration values in each of the three axes shown in Figure 1. https://developer.apple.com/documentation/coremotion/getting_raw_accelerometer_events https://developer.apple.com/documentation/coremotion/getting... Edit: this can get a lot more sophisticated if you can combine data from multiple users in the same area. Let's say that people in the left lane are moving at 40 mph and in the right lane are moving at 30 mph. You can easily refine your guesses. All your left lane guys should be traveling about 10 mph faster than your right lane guys. In fact that info is so widely available that I'd be surprised if Google and Apple weren't already using it.