Self-driving cars are a hot field of research nowadays, but some issues remain unsolved. Current AI may for example have problems with scenarios where a cyclist is temporarily obscured for a few seconds by a car. The AI can then forget that the cyclist is still behind the car and has much less time to react when the cyclist reappears.
Researchers at Örebro University, among others, have developed a technology that, in combination with in-depth learning, gives the AI something similar to common sense, allowing it to realize that the cyclist remains behind the car. This allows the AI to plan for the reappearance of the cyclist.
"Our algorithm gives a self-driving vehicle the ability to understand, in this case, that visibility is temporarily obscured by another car and that the cyclist will again be visible when that car has passed. It is understanding at this level that is required for self-driving cars to be used in traffic, under a variety of traffic conditions and in different environments", says Mehul Bhatt, professor of computer science at Örebro University and one of the researchers behind the technology, in a press release.
Another advantage of the technology is that the AI can explain why it has made a certain decision, such as a sharp deceleration when the person in the car does not see any obstacle on the road.
"It is of the utmost importance that we do not have AI technology that drives us around in traffic and that no one fully understands, not even those who developed it or manufactured the vehicles in which it is used. If self-driving cars are to share space with us humans, we know how they make decisions", says Mehul Bhatt.