You know how toddlers imitate grownups when it comes to taking their first steps or tying their shoelaces? Well, self-driven cars are starting to do the same. Soon enough, more humanized cars will mirror your moves to predict driving reactions.
Researchers from Cornell University have worked on a system that trains a camera to follow drivers’ movements in detail and transform them in valuable data. In this way, autonomous cars will be packed with data surroundings from the road and a natural human response to more subtle threats.
By collecting data from 10 drivers with 1000 covered miles over a period of two months, researchers formed a Recurrent Neural Network to predict what the person behind the wheel will do. The system registered speed, GPS locations and head orientation in order to establish drivers’ response to changing lanes. The results were better than they could have expected; the software “can anticipate maneuvers 3.5 seconds before they occur in realtime”.
Through deep learning, it was proven that cars can learn from humans when to switch lanes… although, we will have to see if they can take it to the next level – overtaking other cars or avoiding reckless drivers.