Tesla makes the most popular electric vehicles on the market today. Many of Tesla’s vehicles are equipped with its Autopilot driver assistance system, allowing vehicles to drive themselves in some situations. However, Autopilot does not make Tesla’s automobiles fully autonomous, and the system requires driver attention.
Unfortunately, several high-profile crashes resulting in fatalities are alleged to have occurred while Autopilot was engaged. A new study conducted by MIT is investigated what it terms non-critical disengagement of Tesla’s Autopilot in naturalistic highway driving. The study set out to determine if drivers are becoming inattentive while using partially automated driving systems.
The main premise of the study is that the safety impact of inattention behind the wheel with systems like Autopilot is unknown until we have data on how visual behavior is changed with automation. The study investigated data from 290 human-initiated Autopilot disengagements. In addition, study researchers investigated glance duration and transition using Bayesian Generalized Linear Mixed models.
In the study, the model replicated observed glance patterns across a range of drivers. Researchers found that off-road glances were longer with Autopilot active than when it was when inactive. The team also found that the frequency of off-road glances specifically related to driving were less frequent with Autopilot active than while manually driving.
During driving, non-driving-related glances down or to the center stack areas were the most frequent and longest, with 22 percent of those glances exceeding two seconds. The study found little difference in on-road glance duration. The study concluded that visual behavior patterns change before and after Autopilot is used. Before Autopilot was disengaged, drivers in the study looked at the road less and focused more on areas unrelated to the driving task than when they were driving manually.