Kakao Mobility announced on the 14th that its technology, which analyzes driver behavior data of the route provided by navigation using artificial intelligence (AI) and reflects it in route guidance, was recognized for its technological prowess by being published in the SCI-level journal 'Transportation Research Part C: Emerging Technologies' earlier this month.
A paper on AI route guidance technology based on driver reactions, co-authored by the Kakao Mobility AI research and development team and Professor Kim Dong-kyu's research team at Seoul National University, gained attention in academia as a pioneering case that reflects the potential characteristics of roads not considered by existing navigation systems in route exploration based on driver behavior data and confirms its effectiveness in actual commercial services.
Navigation systems generally explore routes based on physical information such as vehicle speed and the number of lanes. However, drivers may deviate from the routes provided by navigation for various reasons, such as areas with illegal parking, inconvenient exits and entrances, or distrust of unfamiliar routes. However, reflecting all these factors in route guidance is practically difficult.
Kakao Mobility found a solution by comparing and analyzing the routes guided to drivers with actual driving data. By assessing the 'traffic value' of roads based on the 'route compliance rate,' which sees whether drivers actually drove the guided roads, it was incorporated into route exploration. This was made possible by building an AI algorithm that applies 'Multi-armed Bandit (MaB),' a methodology of reinforcement learning that evaluates the value of specific items based on user responsiveness.
Through this, the system can automatically learn the inconvenient factors that influence drivers' route selection, continuously improving usability without the need for separate infrastructure construction. It has also become possible to more precisely calculate the traffic value of millions of road segments across the country and increase the accuracy and reliability of route guidance by also reflecting real-time traffic information.
Since November of last year, Kakao Mobility has applied driver reaction-based AI route guidance technology to Kakao Navigation. When a driver selects a destination, the Kakao Navigation algorithm applies this technology centered around 'fast route,' 'highway priority route,' and 'major road priority route' to suggest a 'Navigation recommended route.'
The effectiveness of the technology was also published in the paper. According to the paper, an analysis of data during the initial week of technology application showed that the route compliance rate for drivers based on different route exploration methods increased from 64.22% to 70.87% for the 'fast route.' For the 'highway priority route,' it rose from 71.32% to 72.91%, and for the 'major road priority route,' it improved from 70.79% to 72.40%.
Kim Purmye Kakao Mobility AI research and development team researcher, the first author of the paper, noted, "This study focused on quantifying the degree of discrepancy between the information considered by navigation during route exploration based on user behavior data and the actual driving environment, aiming to provide improved routes by reducing the gap," adding, "We confirmed a significant achievement academically and in service by verifying improved effects in various route quality indicators such as actual driving time to the destination and driving convenience of the road."