Professor Uichin Lee from the Department of Industrial and Systems Engineering has led a team of researchers in developing an AI software that can evaluate whether it is safe for a driver’s digital assistant to provide auditory-verbal services.

Digital assistants in vehicles began as touch screens near the dashboard, which required the use of drivers’ eyes and fingers to interact. These soon developed into auditory-verbal systems, similar to Apple’s Siri, providing services in the form of conversation so that drivers can keep their eyes on the road and hands on the wheel. However, studies show that these still distract drivers enough to reduce their attention to the road and driving safety.

For such services to be as safe as possible, their interactions with the driver must be kept to a minimum. However, to maximize quality, they must provide all needed information at any desired time. The goal of Professor Lee’s team, which includes PhD students Auk Kim and Woohyeok Choi from the Graduate School of Knowledge Service Engineering, was to develop an AI that will consider the safety and necessity of providing auditory-verbal services at any given moment.

They developed an AI with a composite cognitive model that considers two factors: interruptibility and necessity. For the former, they integrated their system to a navigation app and allowed it to collect real-time data inside the car (e.g. brake pedal position, steering wheel angle) and outside (e.g. distance between cars, flow of traffic) to determine if a given moment is safe to interrupt. Then, each interaction’s necessity is evaluated — for instance, an incoming call would be more necessary than a simple weather update. After conducting a real-road field study with 29 drivers, they found their AI to be at most 87% accurate in determining “interruptible” moments.

Regarding the benefits of their innovation, Professor Lee said that it “will certainly help vehicles interact with their drivers safely as it can fairly accurately determine when to provide conversation services using only basic sensor data generated by cars”. Such technology, further improved, will be valuable in the integration of digital assistants into vehicles.