Developed artificial intelligence will spread autonomous vehicles

A joint effort between the Universities of Boğaziçi and Oxford has developed a new artificial intelligence that will enable autonomous vehicles to drive more safely in poor weather conditions. This has overcome one of the biggest obstacles to the spread of self-driving vehicles (AVs). With the artificial intelligence developed, the vehicles can handle difficult conditions such as snow, rain, fog and night vision, making them safer.

In about two and a half years of collaboration between the Universities of Boğaziçi and Oxford, scientists have developed an artificial intelligence architecture that is crucial for autonomous vehicles to enable safe autonomous driving, especially in difficult weather conditions and in environments without GPS. The study, published in Nature Machine Intelligence, one of the world's most prestigious scientific journals in the field of machine learning, added to the literature a self-supervised deep learning architecture for self-motion prediction, which is a critical part of the software core of autonomous vehicles. The research applied an intelligent sensor fusion method with geometric awareness to leverage the strengths of multiple sensors such as camera, radar, and lidar.



Dr. Mehmet Turan, a lecturer at the Faculty of Computer Engineering at Boğaziçi University, who was involved in artificial intelligence technology in the study, explained that autonomous vehicles are cyber-physical systems that perform very complex functions such as mapping, positioning, navigation, route calculation and implementation, adding, "In recent years, data-based algorithms promise to minimise vulnerability to environmental factors. However, in real-world conditions, situations that reduce the performance of this modular structure to critical levels that can endanger driving safety may arise, and this is one of the biggest obstacles to the widespread use of autonomous vehicles in everyday life. In this context, one of the most critical software modules in autonomous vehicles are precision positioning systems. These systems form the basis for important functions such as motion planning, prediction, situational awareness, and collision avoidance in the core software, which we can call the brain of the autonomous vehicles. The artificial intelligence we have developed combines sensors such as camera, radar, and lidar in real time and effectively according to road conditions, so that autonomous driving is possible even in difficult weather conditions. The artificial intelligence model we developed can decide which sensors should be trusted more and which less in difficult weather conditions such as snow, rain, fog, or night. Thus, the system is able to maintain its precise positioning even in difficult conditions."


Yasin Almalıoğlu, Professor of computer science at the University of Oxford, on the other hand, pointed out that the artificial intelligence developed is a software that is crucial to increasing the safety and reliability of autonomous vehicles in bad weather conditions, and continued as follows "This software enables autonomous vehicles to determine their position precisely in difficult conditions. To achieve this, it interprets the various sensor information in the most effective way depending on weather and road conditions. Thus, the sensor performs fusion. To understand the importance of precise positioning, imagine an autonomous vehicle that stops a few meters before a traffic intersection or misjudges the lane before turning.

It is not difficult to predict that many accidents with fatal consequences will occur in such and similar situations. If we consider how many lives these and similar accidents may cost per day as the use of autonomous cars becomes widespread in the future, we will better understand the importance of precise positioning technologies."

Dr. Mehmet Turan and Dr. Yasin Almalıoğlu agree that this study makes autonomous vehicles safer, takes a step closer to operating in all weather conditions, and will have a decisive impact on social autonomous vehicle applications.

Click here to access the study published in Nature Machine Intelligence.


Çeviren:Duygu Melek Anık