Kakao Mobility honing self-driving tech to bring robotaxi to everyday use
Summary
Kakao Mobility is honing end-to-end self-driving technology to bring robotaxi service into everyday use in Korea. The company has been testing in Seoul’s Gangnam area since March and says the pilot has produced more than 13,000 kilometers of driving data. It plans to use a single E2E model for perception and decision-making by the end of this year, while keeping rule-based safety checks. The company said the pilot has recorded about 2,000 rides, an average customer rating of 4.97, and one minor rear-end accident with no injuries.
Key Facts
- Kakao Mobility launched an autonomous passenger transport service in Gangnam in March this year, operating from 10 p.m. to 5 a.m.
- The service began with two vehicles and expanded to six in August.
- The pilot fleet has logged more than 13,000 kilometers and completed about 2,000 rides.
- Kakao Mobility said the service has received an average customer rating of 4.97 out of 5.
- The company said there was one minor accident in which another vehicle rear-ended the self-driving car, but no injuries were reported.
Kakao Mobility's self-driving test vehicle is showcased at a press briefing at the Korea Automobile Manufacturers Association headquarters in southern Seoul, Wednesday. Korea Times photo by Nam Hyun-woo
Kakao Mobility is honing its end-to-end (E2E) self-driving technology to bring self-driving taxis into everyday use in Korea, building on driving data accumulated over the past six months through an ongoing pilot program in Seoul’s bustling Gangnam area.
The ride-hailing platform held a press briefing on Wednesday, showcasing its latest progress in self-driving services and outlining its autonomous driving technology and integrated platform strategy.
Kakao Mobility formed its autonomous driving team in 2018 and began testing the technology in 2020. In March this year, it launched an autonomous passenger transport service in Seoul’s Gangnam area, operating from 10 p.m. to 5 a.m.
The service started with two vehicles and expanded to six in August. Kakao Mobility said the fleet has logged more than 13,000 kilometers so far, completed about 2,000 rides and received an average customer rating of 4.97 out of 5.
Kakao Mobility’s pilot program seeks to operate a mixture of E2E technology and a rule-based model.
A rule-based model divides autonomous driving into separate modules for sensing, perception, planning and control, with the vehicle following rules defined by human developers. The approach makes it possible to verify how the system reaches its decisions, but requires developers to anticipate virtually every situation that could occur on the road, which is a nearly impossible task.
On the other hand, an E2E model processes sensor data through a single artificial intelligence (AI) model and directly translates it into vehicle control without defining the intermediate steps. While a rule-based system is like feeding the vehicle a driving manual line by line, an E2E model is more akin to letting it learn by observing millions of hours of driving.
“If you ask a driver with 20 years of experience why they made a certain decision in a particular situation, they may simply say, ‘It just felt like the right thing to do,’” said Im In-ho, head of Kakao Mobility’s self-driving development team. “Learning that kind of judgment is what E2E is about.”
Kakao Mobility currently uses both rule-based and AI planners at the planning stage, where the vehicle decides how to maneuver. By the end of this year, the company plans to use a single E2E model for both perception and decision-making, while retaining rule-based safety checks to verify its decisions.
Im In-ho, head of Kakao Mobility’s self-driving development team, speaks during a press briefing at the Korea Automobile Manufacturers Association headquarters in southern Seoul, Wednesday. Korea Times photo by Nam Hyun-woo
Im said Kakao Mobility chose Gangnam because the area provides various “edge cases,” referring to unexpected situations encountered while driving. Dashcam footage shown during the briefing captured situations including vehicles making illegal U-turns, cars crossing the center line because of maintenance work, drunk pedestrians suddenly entering the road and cargo falling from delivery trucks.
“Driving for one month in Gangnam can expose us to far more unexpected situations than driving for a year in an easier environment, which has helped us secure valuable data,” Im said.
Kim Min-sun, leader of the company’s self-driving business team, said there had been one minor car accident in which another vehicle rear-ended the self-driving car, but no accidents involving injuries during the pilot period.
She said Kakao Mobility will use its capabilities not only in autonomous driving technology but also in platform operations to improve the service.
Explore More
- Q.
- Q.
- Q.