MIT study: Tesla drivers become inattentive when the autopilot is on

By the end of this week, thousands of Tesla owners may be testing the latest version of the Full Self-Driving (FSD) beta software, version 10.0.1, on public roads, even as regulators and federal officials keep safety out after a few high-profile accidents examine the system.

A new study by the Massachusetts Institute of Technology supports the assumption that the FSD system, which despite its name is actually not an autonomous system, but rather an advanced driver assistance system (ADAS), may not be that safe at all. Researchers who examined the gaze data of 290 human-initiated autopilot shutdowns found that drivers can become inattentive when using partially automated driving systems.

"Visual behavior patterns change before and after the [autopilot] deactivation," says the study. "Before the deactivation, the drivers looked less at the road and concentrated more on non-driving-relevant areas than after the transition to manual driving. The higher proportion of glances away from the road before the switch to manual driving was not compensated for by longer looks forward . "

Tesla CEO Elon Musk has stated that not everyone who has paid for the FSD software will have access to the beta version, which promises more automated driving functions. Initially, Tesla will use telemetry to collect personal driving data over a seven-day period to ensure drivers are still paying enough attention. The data can also be used to set up a new safety rating page that tracks the owner's vehicle and is linked to their insurance.

The MIT study provides evidence that drivers may not use Tesla's autopilot (AP) as recommended. Because the autopilot includes safety features such as traffic-sensitive cruise control and auto-steering, drivers are less alert and take their hands off the wheel more often. The researchers found that this behavior may be due to drivers not properly assessing the capabilities and limitations of the AP functions, which is compounded when they perform well. Of course, drivers whose tasks are automated for them can get bored after trying to maintain their visual and physical alertness, which, according to the researchers, only leads to further inattentiveness.

The report, entitled "A model for naturalistic glance behavior around Tesla Autopilot disengagements" has Tesla Model S and Model X owners during their daily routine for a period of observed in the greater Boston area for a year or more. The vehicles were equipped with the Real-time Intelligent Driving Environment Recording1 data acquisition system, which continuously collects data from the CAN bus, a GPS and three 720p video cameras. These sensors provide information about the vehicle kinematics, the interaction of the driver with the vehicle control units, the mileage, the location and posture of the driver, his face and the view in front of the vehicle. MIT has collected nearly 500,000 miles of data.

The aim of this study is not to put Tesla to shame, but rather to advocate systems for controlling driver attentiveness that give the driver feedback in real time or that can adapt the automation functions to the level of driver awareness. Currently, autopilot uses a hand wheel sensor system to monitor driver engagement, but it does not monitor driver attention through eye or head tracking.

The researchers behind the study developed a gaze behavior model, "based on naturalistic data, that can help understand the characteristics of shifts in driver attentiveness under automation and support the development of solutions to ensure that drivers are safe." stay sufficiently busy with the driving tasks. " Not only would this help driver monitoring systems cope with "atypical" looks, but can also be used as a benchmark for studying the safety effects of automation on driver behavior.

Companies like Seeing Machines and Smart Eye are already working with automakers like General Motors, Mercedes-Benz and allegedly Ford to develop camera-based driver monitoring systems for vehicles with ADAS, but also to solve problems caused by drunk driving or inability to drive. The technology is there. The only question is: will Tesla use them?