Seems like they must require a “keep alive” (dead man’s switch) to ensure driver’s remain alert.
Among other things, the office wants Tesla to improve its driver engagement monitoring systems. Tesla relies on sensing a driver’s hands on the steering wheel to know if they are attentive enough to the road while using Autopilot features. Other auto makers use cameras to ascertain whether a driver’s eyes are on the road.
I understand train locomotives require that the engineer make at least one control input every 25 seconds. If that doesn’t happen, the train will begin braking. Seems like pseudo-self driving cars must use a similar approach – if the driver does not manipulate the controls, then alerts must be activated and/or the car needs to get off the highway and stop.