Artificial intelligence from MIT predicts what drivers you are

Share:

New technology for autonomous cars.

Although autonomous cars are rare on the roads so far, their numbers are slowly but steadily increasing, especially in countries such as the USA. This fact can be very problematic, and not because autonomous cars can take work from taxi drivers, for example, and because so far the coexistence of traditional and autonomous cars can pose a threat to road users. This threat is due to the fact that artificial intelligence systems tend to assume that each driver behaves in the same rational and predictable way. If you have a driving license, you certainly know that this is far from the truth.

The problem is currently being resolved with researchers from the CSAIL ( Computer Science and Artificial Intelligence Laboratory) MYTH. As anticipating driver behavior is to be extremely difficult for artificial intelligence (because it requires a certain degree of social awareness), scientists have used tools in the field of social psychology. With the help of these tools, they helped the system classify driver behaviors, dividing these behaviors into selfish and altruistic.

When MIT researchers tested their algorithm, with the help of maneuvers involving lane change and dangerous left turns, it turned out to be able to predict the behavior of other cars with an accuracy 25 percent higher than before. During the left turn simulation, a scholarly car knew, for example, that he had to wait when the "selfish" driver drove from the opposite direction, and that he could turn when the "disinterested" driver drove from the opposite direction.

myth si
Artificial intelligence from MIT can determine whether the driver in the car in front of him is a driver who exhibits altruistic or egoistic behavior. Source: MIT

"Working with and around people means having to understand their intentions to better understand their behavior." said the main author of the scientific work describing the algorithm, Wilko Schwarting. "People's tendencies to cooperate or compete often affect how drivers behave. When writing this paper, we tried to understand if it could be quantified. "

Of course, before the algorithm can be used on the roads, it must be thoroughly refined. However, this does not change the fact that this system can be very useful, and not just for self-driving cars. This system could also generate hints for drivers themselves, helping them to adapt their behavior to the behavior of drivers around them. In addition, it could help the autonomous automate to display more human behavior, which would make it easier for drivers to understand the behavior of these vehicles on the road.

Source: MYTH