Seoul Autonomous Vehicle near Maebong Station in Gangnam-gu, Seoul, on the 31st last month./Courtesy of Lee Hojun

In the early days of operations, there were cases where the vehicle shook or moved unstably in tunnels, but as we improved perception and positioning performance, it can now pass through without any wobble.

At 11 p.m. on the 31st, a tunnel in Gangnam District, Seoul. The GPS that had been showing the current location on a smartphone map app temporarily stopped. It was because the vehicle entered a shadow zone where satellite signals cannot reach. Still, the autonomous vehicle maintained speed stably and exited the tunnel. The safety operator in the driver's seat did not touch the wheel. A Kakao Mobility official said, "When GNSS (Global Navigation Satellite System) is not available, we determine the vehicle's position through sensor fusion using lidar, cameras, and radar."

Kakao Mobility's Autonomous Driving service "Seoul Autonomous Car" is evolving after running more than 2,000 times through downtown Gangnam. Based on data accumulated over about six months since operations began in Mar., the number of lidars was reduced from five to three, and safety operator interventions also fell sharply. It drives stably even in tunnels where GPS is cut off, and since last month it has expanded its operating range to include protected zones for children, older adults, and people with disabilities. However, expensive sensors, responses to bad weather, and limits on operating areas and hours remain challenges to solve for level 4 Autonomous Driving commercialization.

◇ Learning from 2,000 runs… lidar cut from 5 to 3

Seoul Autonomous Car is built on Kia's EV6 electric vehicle. What stood out was the in-house sensor kit "AV-Kit," which integrates various sensors such as lidar, radar, and cameras on the roof of a standard EV6. Including the kit, there are seven cameras in total around the front, rear, and sides of the vehicle, five radars, and three lidars.

According to Kakao Mobility, Seoul Autonomous Car perceives 360 degrees around the vehicle through a hybrid perception system that uses lidar, radar, and cameras together. In particular, AV-Kit is designed so that the types of sensors and mounting positions can be changed, making it installable on other cars.

Kakao Mobility is developing not only sensors but also the driving artificial intelligence (AI) and the data learning framework in-house. It improves performance by reusing driving data collected on real roads for AI training.

Gangnam District is one of the areas with a high Autonomous Driving difficulty due to frequent cut-ins, illegal parking and stopping, and jaywalking. Seoul Autonomous Car repeatedly runs here to collect "edge cases," unexpected situations outside the norm. Since the service began in Mar., completed trips have exceeded 2,000, with more than 10,000 kilometers in cumulative mileage. According to the company, more than gigabytes (GB) of data are accumulated from just one hour of driving.

Seoul Autonomous Vehicle's in-house sensor kit AV-Kit./Courtesy of Lee Hojun

As Autonomous Driving technology has advanced, the number of sensors needed in the vehicle has decreased. The vehicle deployed first in Mar. carried five lidars, but the newly added vehicle has only three. Given that the price of the Autonomous Driving equipment currently installed in Seoul Autonomous Car alone exceeds the price of one EV6, sensor optimization is a key task for commercialization.

A Kakao Mobility official said, "To scale up Autonomous Driving services, we need to enhance perception performance while reducing the number of expensive sensors to lower costs."

Even with fewer sensors, driving performance has improved. According to Kakao Mobility, compared with the early days of operation, the number of times a safety operator intervenes has dropped significantly, and recently there have been many trips completed without a single intervention. During the roughly 30-minute test ride that day, the safety operator did not take the wheel.

User feedback is also positive. The average rating from users who evaluated the Seoul Autonomous Car service was 4.97 out of 5, and the company said many passengers noted they were surprised by higher-than-expected Autonomous Driving performance after actually riding. There were no accidents with reported bodily injury or property damage over the past six months, and in the single accident that occurred, the following vehicle rear-ended Seoul Autonomous Car.

As Seoul Autonomous Vehicle enters a school zone, the road on the passenger monitor turns red, while the blue line indicates the vehicle's expected route./Courtesy of Lee Hojun

◇ Speed drops to 24 kph upon entering a child protection zone… normal operations up to 5 mm of rainfall per hour

Since the 20th of last month, Seoul Autonomous Car has increased the number of vehicles from two to six. At the same time, it expanded its Autonomous Driving range to include protected zones for children, older adults, and people with disabilities. Until now, due to related regulations, the Autonomous Driving function could not be used in these zones, and safety operators had to drive directly.

When the Seoul Autonomous Car boarded that day entered a child protection zone, the road displayed on the rear-seat monitor turned red. The vehicle's speed also dropped noticeably. Kakao Mobility applied a separate safety control algorithm to ensure that in protected zones for transportation-vulnerable people, the vehicle travels at 80% or less of the speed limit. In zones with a 30 kph limit, the maximum is 24 kph, and in 50 kph zones, the maximum is 40 kph. The company explained that once enough driving data in protected zones is accumulated and no particular safety issues arise, it plans to consider gradually adjusting the speed criteria.

Although 1 to 2 mm of rain per hour fell in Gangnam that day, there were no issues with operations. Kakao Mobility, for safety, temporarily suspends Seoul Autonomous Car operations when hourly rainfall exceeds 5 mm. However, the company said that even if rainfall exceeds 5 mm per hour during a trip, the vehicle will continue to the destination.

◇ Sensor perception in worsening weather remains a task… only six vehicles

Although Autonomous Driving technology is advancing rapidly, there are still challenges to resolve. The vehicles are currently designed with sensor ventilation structures to prevent moisture buildup, but plans are underway to enhance them by introducing separate cleaning devices that automatically remove water droplets or foreign substances on sensors to ensure camera and lidar perception even in heavy rain or snow. The ride quality was not smooth in every section. The Seoul Autonomous Car boarded that day also came to an abrupt stop when the light turned red just before a right turn.

The scale of operations still does not meet demand. Even though the number of vehicles was tripled from two to six, it is not easy to hail a Seoul Autonomous Car at the desired time. According to the company, vehicle utilization is high at over 80%, leading to a fair number of failed hails. Operating hours are limited from 10 p.m. to 5 a.m. the next day, and the service area remains confined to Gangnam District, both of which must be addressed to expand the service.

Kakao Mobility is considering increasing operating hours, areas, and the number of vehicles. Since Jun., it has been collecting driving data in Gangnam during daytime hours. The company said that because there are more vehicles and pedestrians and more frequent unexpected situations, the technical difficulty of daytime Autonomous Driving in Gangnam is higher than at night.

Expanding the operating area beyond Gangnam is also under review. A Kakao Mobility official said, "We are considering areas that need service expansion after adding vehicles, not only Gangnam but also Seocho and Songpa," and added, "As we have recently released cooperation with several original equipment manufacturers (OEMs), we are also reviewing the development of new models."

※ This article has been translated by AI. Share your feedback here.