How to determine the ratio between false positives (driver is incorrectly detected to be inattentive/drowsy) and false negatives (driver is incorrectly not detected to be inattentive/drowsy)? Too many false negatives may result in safety risks, too many false positives in reduced trust and acceptance.
Should we aim for detecting a required driver state (given the driving conditions such as traffic complexity and environmental conditions) or are the driver state criteria for transition of control always the same (absence of distraction and drowsiness)?
What kind of misuse and abuse do we foresee? How to prevent them?
If driver monitoring becomes an essential element of transition of control, how do we deal with driver diversity (age, size, skin colour, etc)? Will automated cars work better for some people than others?
HMI for driver monitoring systems: what should the user know about what the vehicle monitors and what driver state the vehicle estimates?