LG Innotek’s autonomous-driving data collection vehicle travels on a city street in Korea. (LG Innotek)
LG Innotek’s autonomous-driving data collection vehicle travels on a city street in Korea. (LG Innotek)

LG Innotek has begun collecting real-world driving data on Korean roads with US software company Applied Intuition to improve sensors and software for autonomous vehicles, the company said Tuesday.

The first vehicle will soon start operating in Seoul, Incheon and Gyeonggi Province under a partnership formed in March. LG Innotek plans to expand the Korean fleet to about 20 vehicles for data collection and autonomous-driving software development by 2028.

Equipped with cameras, radar and lidar, the vehicle will record road conditions, traffic movements, object locations and weather. The companies will combine the data with digital simulations of real-world environments to test individual sensors and how they work together.

A central focus is sensor fusion: combining camera images, radar measurements of distance and speed, and lidar-generated 3D maps to help vehicles interpret their surroundings more accurately.

Testing will also cover difficult driving conditions, such as detecting black vehicles in heavy rain, navigating strong backlighting at tunnel exits and reading irregular lane markings at construction sites. These scenarios can reveal weaknesses that routine tests may miss.

Applied Intuition also plans to deploy vehicles equipped with LG Innotek sensing modules in the US, Europe and Japan. Data from different road environments will help identify performance issues and refine the modules.

Pedestrian faces, license plates and other personal information will be anonymized before the data is used, LG Innotek said.

“Autonomous-driving solutions are evolving toward jointly optimizing sensors and software based on data,” said John Min, chief technology officer at LG Innotek. “We will accelerate our push into the physical AI sensing market with advanced multi-sensor solutions.”

By Lim Ye-jin (yejinlim@heraldcorp.com)