Blog Page


Boğaziçi, Oxford develop AI for safety of autonomous vehicles | Daily Sabah – Daily Sabah

Boğaziçi University has developed and trained artificial intelligence (AI) models for the challenging and complex scenarios that vehicles may encounter on the road under all kinds of unsuitable conditions.
In this regard, studies are being conducted under the leadership of Boğaziçi University Computer Engineering faculty member Dr. Mehmet Turan in cooperation with Oxford University.

Explaining that AI has come a long way in recent years, Turan noted that there have been many interesting developments in the field, especially during the last 10 years. “Even in the most advanced form, AI is still far behind the functioning of the human brain,” he added.
Considering the high-level functionality of human intelligence such as reasoning, analysis, deduction, decision making, or its ability to produce solutions for situations that the brain has not encountered before, Turan said the concept of fully autonomous driving that humans have been chasing for years is still a far-fetched dream.

“I think there is a low probability of these dreams to come true in the next 10-15 years,” he opined, adding that the tech giants such as Tesla and Google are conducting academic and industrial studies for fully autonomous vehicles but on relatively small-scale trial programs.
“Imagine a pedestrian suddenly appears on the road, or the road is blocked due to some renovation, or two vehicles bump into each other posing a danger to the vehicle behind. Think about the shortening of the sensors' vision distance in this scenario. The artificial intelligence we have developed will rely more on the data from the sensors,” Turan explained.
We are working on artificial intelligence models especially developed and trained for challenging and complex scenarios so that autonomous vehicles can drive safely on all kinds of roads and in all kinds of weather conditions, he added.
“When the models encounter such challenging scenarios, the solutions applied by experienced drivers are artificially produced, accompanied by supervised learning, focusing more on artificial intelligence algorithms that are built considering various observations,” Turan concluded.