You could do that, theoretically, from several days of driving, with false positives and false negatives galore. Or you could ask the user...I know, not data driven, boooriiiing, and requires one more interaction.

How would that be possible? My sister has a new VW Polo with a 1.0L 65bhp engine, the official time to 60 is a glacial 16.5 seconds. I'm sure its acceleration profile would match that of a fully loaded HGV.

The time to turn, the time to break, way of turning lanes, all of it could be used to infer a vehicle type. I am not saying they do it, but it's likely.

If this information were used to infer vehicle size, Google would assume everyone in Seattle is driving an 18 wheeler.

It's quite error prone. I doubt they do it. Also deriving acceleration data requires sub second speed sampling. This needs to be done on device as it sends too much data sampling over the wire otherwise.

Can't imagine that being of any use in London, nothing moves.

Or more easily from the default Bluetooth name