They have discovered that through the use of certain iPhone sensors, like the built-in accelerometer, they are able to capture certain traits of the user that might be an indicator that they’re too drunk to drive. This is based on the gait of the user, where the researchers found that there are certain gaits that are unique to people when they are drunk.
Based on this, they tested out their theory and found that in their tests, they managed to achieve a 90% accuracy rating, which is actually pretty good. However, before you get too excited, the researchers note that this is more of a proof of concept study as opposed to an actual development of an app.
The test was also conducted under controlled conditions, and the researchers note that this is based on users putting their phones on their lower backs, like in a backpocket or a bag, meaning that it might not be as accurate if the person is holding their phone or if it is placed somewhere else on their person.