Researchers will use deep-learning algorithms to study video feeds of the driver and forward roadway, along with data streams from the vehicle's automated systems.
A group of researchers from the Massachusetts Institute of Technology have launched a large-scale study of driver interactions with vehicle automation systems, requiring dozens of vehicles to be equipped with driver-facing cameras and data logging equipment.
Some automakers see full Level 4/5 autonomy on the horizon, while others caution that current software is still several years away from safely taking full control of a vehicle as the driver takes a nap.
The MIT team is focused on better understanding human-machine interactions as automated systems become more capable yet require a human driver to retain control over certain tasks or remain attentive and able to intervene in certain situations.
The study has already modified more than two dozen vehicles, mostly the Model S with a few wild cards such as the Range Rover Evoque and Volvo S90. Each has been equipped with driver- and forward-facing video cameras and data logging gear to capture IMU, GPS and CAN messages.
So far, the study has already involved 78 participants over more than 275,000 miles.
Machine vision algorithms are used to process the video recordings, creating tons of data to help study correlations between the driver's body position and gaze, the driving scene, weather and the vehicle's behavior.
"Analyzing the long-tail data requires processing billions of high-definition video frames with state-of-the-art computer vision algorithms multiple times as we learn both what to look for and how to interpret what we find," the authors write (PDF).