This excerpt taken from MacStories Federico Viticci in his iPhone X review, nails it:
“Face ID builds upon the more capable face recognitions APIs introduced in iOS 11 through a dedicated, self-contained, self-learning process that uses errors as data points for improvement. According to Apple, when Face ID fails and you have to enter your passcode manually, if the data recognized by Face ID hits a certain threshold of familiarity with the user, the passcode is used as confirmation to add the unrecognized scan to Face ID’s data pool and, hopefully, teach it to recognize a similar scan in the future.”
The biggest fear I had prior to purchasing the X was not being able to unlock the phone in situations where using TouchID made it convenient. Laying down in bed, with the phone laying flat on the desk or mounted in my car using the CD mount.
After a month of using the X and allowing machine learning to get used to the daily operation of my device, I find that even when I have half my face in the pillow, it unlocks. When I am driving in the car the phone now recognizes me, even if I am off on a slight angle from the phone.
Remember that each time FaceID fails, using the passcode to unlock the phone adds that unrecognized "angle" to the data pool for FaceID.
At this point I do not miss TouchID at all.