2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
Map of the movement of vehicles in the study in New England
Since the beginning of 2016, the Massachusetts Institute of Technology has been conducting a large experiment to collect data on the interaction of drivers with semi-automatic control systems in cars. For the first time, the authors of the experiment revealed the details of the study, its methodology and intermediate statistics, according to Wired.
Many major automakers are involved in prototyping and testing self-driving cars on landfills and on public roads, sometimes even without a tester driver behind the wheel. As a rule, during such tests, the emphasis is on the technical aspects of the trip and the adequacy of the decisions of the control algorithms, and not much attention is paid to the interaction of drivers and passengers with such systems.
In 2015, MIT, together with several companies, including automakers, founded the Advanced Vehicle Technology consortium, whose goal is to study how people interact with new technologies in cars, including driver assistance systems and autopilots. Now researchers led by Lex Fridman and Bryan Reimer at MIT have revealed details about the study, which began in January 2016.
In order to obtain data as close as possible to the actual use of cars, the researchers decided to collect data from each driver over medium (month) and long (a year or more) periods of time, and without hard restrictions. Thus, the drivers used cars for their regular journeys, rather than moving on them along strictly defined routes.
To collect data, the researchers installed cameras in cars that aimed at the driver's face, the entire driver's compartment, the road, and in some cases, the dashboard. In addition, the researchers collected data from the car's CAN bus, a gyro stabilizer, and a GPS receiver using a single-board computer specially installed in each car.
Cars that have participated or have previously participated in the experiment, as well as intermediate research statistics
As of October 23, 2017, 78 drivers and 25 vehicles took part in the study. During the experiment, the cars drove almost 450 thousand kilometers, and the cameras installed on them filmed about three and a half billion frames. Since this amount of data is difficult to process manually, the researchers decided to use deep machine learning for this. With its help, they were able to automatically receive data from the "raw" data about the driver's condition, his posture, the state of the car and the environment, divided into segments corresponding to objects of different types (road, sidewalk, pedestrians, etc.).
The exact timing of the study is unknown. The researchers believe that the data accumulated during its conduct will help manufacturers of fully or partially autonomous cars to better understand people's attitudes towards certain aspects of their use.
Other companies and research organizations are also studying the interaction between people and self-driving cars. For example, the Russian company Cognitive Technologies, which is involved in the development of unmanned vehicles, conducted a sociological study to identify the moral criteria for the system's behavior in critical situations, and later the results of a similar international study were published.
In addition, Ford has recently tested a self-driving car system that allows them to communicate their next maneuvers to those around them. Interestingly, the company initially began research without public announcements. It became known about it only after residents of the American city in which the testing was carried out began to notice a car moving around the city without a driver and passengers, and then they found out that in fact a man dressed in a car seat suit was sitting at the wheel of the car.