MIT Study: Tesla Autopilot Leads to Driver Inattention

Tesla’s Full Self-Driving Beta Under Scrutiny

Potentially thousands of Tesla vehicle owners are poised to participate in testing the latest iteration – version 10.0.1 – of the automaker’s Full Self-Driving (FSD) beta software on public roadways this week. This rollout occurs amidst ongoing investigations by regulators and federal authorities concerning the system’s safety, prompted by several notable accidents.

MIT Study Highlights Driver Inattentiveness

Recent research conducted by the Massachusetts Institute of Technology supports concerns regarding the safety of the FSD system. Despite its designation, the system functions as an advanced driver assist system (ADAS) and is not truly autonomous. The study focused on analyzing glance data collected during 290 instances of human disengagement from Autopilot.

Researchers discovered that drivers exhibited a tendency towards inattentiveness while utilizing partially automated driving systems. “Driver visual patterns demonstrably shift before and after Autopilot disengagement,” the study details.

Changes in Visual Focus

Before disengaging Autopilot, drivers directed their gaze away from the road more frequently, concentrating instead on non-driving related areas. This shift wasn’t offset by increased forward-looking glances.

This indicates a potential safety risk as drivers may be less prepared to regain control of the vehicle when needed.

Access to Beta Limited by Attentiveness

Elon Musk, CEO of Tesla, has indicated that access to the beta version of the FSD software won’t be universally granted to all purchasers. Tesla intends to leverage telemetry data to assess individual driving habits over a seven-day period.

This data collection aims to verify that drivers maintain sufficient attentiveness while using the system. The collected information may also contribute to the development of a new safety rating system linked to the owner’s insurance profile.

Misunderstanding Autopilot Capabilities

The MIT study suggests that drivers may not be employing Tesla’s Autopilot (AP) as intended. Features like traffic-aware cruise control and autosteering can lead to reduced driver attention and a tendency to remove hands from the steering wheel.

Researchers attribute this behavior to a potential misunderstanding of the AP features’ capabilities and limitations, which is often reinforced by successful operation. Automated tasks can induce boredom, diminishing sustained visual and physical alertness, ultimately exacerbating inattentiveness.

Detailed Data Collection Methodology

The research, formally titled “A model for naturalistic glance behavior around Tesla Autopilot disengagements,” involved monitoring Tesla Model S and X owners during their regular commutes in the greater Boston area for a year or more.

Vehicles were equipped with the Real-time Intelligent Driving Environment Recording system, which continuously gathered data from the CAN bus, GPS, and three 720p video cameras. This instrumentation provided insights into vehicle dynamics, driver interactions, mileage, location, and driver posture, facial expressions, and the forward view.

MIT accumulated nearly 500,000 miles of driving data throughout the study.

Focus on Driver Attention Management

The study’s objective isn’t to criticize Tesla, but to champion the implementation of driver attention management systems. These systems could provide real-time feedback to drivers or dynamically adjust automation functionality based on their attentiveness levels.

Currently, Autopilot relies on a hands-on-wheel sensing system to monitor driver engagement, but lacks the capability to track driver attention through eye or head-tracking technology.

Developing a Glance Behavior Model

The researchers have created a model for glance behavior, “based on naturalistic data, that can help understand the characteristics of shifts in driver attention under automation and support the development of solutions to ensure that drivers remain sufficiently engaged in the driving tasks.”

This model can aid driver monitoring systems in identifying “atypical” glances and serve as a benchmark for evaluating the safety implications of automation on driver behavior.

Existing Driver Monitoring Technologies

Companies such as Seeing Machines and Smart Eye are already collaborating with automakers like General Motors, Mercedes-Benz, and reportedly Ford to integrate camera-based driver monitoring systems into vehicles equipped with ADAS.

These technologies also address issues related to impaired or intoxicated driving. The technology is readily available; the crucial question remains whether Tesla will adopt it.