Here's what I mean. Humans have anxiety when they drive, if we can detect the "anxiety level" of a specific driver then we can better predict how to drive with them. Driving ing on the road is a team sport, the better you know your teammates the better the whole team (cars on the road with you) works. Any info on this topic would be super appreciated!
Looking at "Driver Style Recognition" on google scholar seems to return some results including a few GM patents. The papers appear to mostly focus on accelerometer data over longer periods of time, so they don't seem to be directly relevant to labeling the behavior of other nearby cars. There might however be some newer extensions of the work which could be more relevant to self driving cars, though it wouldn't be unsurprising if it was not published information.