Kakao's Late-Night Robotaxi Test in Seoul's Gangnam District

Why Gangnam After Midnight Is the Hard Test

Kakao Mobility — the ride-hailing and mobility arm of Kakao, the company behind Korea's dominant messaging app — is running autonomous vehicles through Seoul's Gangnam district on an unusual schedule: from 11pm to 5am. The details were presented on September 9 at the Korea Automobile & Mobility Association (KAMA) building in Seoul's Seocho district.

Choosing late-night Gangnam over quiet daytime roads is a deliberate engineering call. Gangnam is Seoul's densest commercial and nightlife area, and after midnight it produces exactly the situations rule-based driving software struggles with: illegal U-turns, sudden jaywalking, and intoxicated pedestrians moving unpredictably. A phrase from the event — "even drunk pedestrians are an asset" — captures the view that these edge cases are the most valuable raw material for improving autonomous software.

AV-Kit Plus Round-the-Clock Remote Monitoring

The vehicles carry Kakao Mobility's in-house modular autonomous driving kit, the "AV-Kit," which uses multiple sensor types to perceive the driving environment from several angles. The modular design matters commercially: a kit that can be transplanted across vehicle models scales faster than hardware engineered into one specific platform.

Just as notable is the pairing with 24-hour remote monitoring. Until true driverless operation arrives, the practical safety layer in any robotaxi service is a control center where humans can intervene. By holding both the driving stack and the dispatch/monitoring platform in-house, Kakao Mobility is positioning itself differently from pure technology developers on one side and pure hailing platforms on the other.

Reading the 13,000 km Figure

Reports put Kakao Mobility's autonomous mileage at 13,000 km over five months, with the company pushing toward Level 4 — driving that requires no human intervention within its operating domain. In absolute terms this is modest next to global leaders that describe cumulative testing in the tens of millions of miles.

That makes the difficulty of the route more interesting than the raw distance. Kilometers accumulated in a high-variance urban core at night are not equivalent to kilometers looped on simple repeatable stretches. How much that translates into better software depends entirely on how the data is used, and it would be premature to treat the service as fully validated.

What Actually Changes for Riders

For now, very little changes for the average commuter: the operating window is narrow and the area is limited. But late night is precisely when Seoul's taxi supply shortage has been a recurring complaint, so the test window overlaps with the demand pocket a commercial robotaxi service would target first. The alignment between a real late-night supply gap and the pilot's hours looks strategic rather than accidental.

For the industry, this reads as a shift in Korea's autonomous driving race from technology demos toward operational capability. Service quality will hinge less on whether a car drives smoothly and more on how quickly someone intervenes and re-routes dispatch when an edge case appears — likely a big reason Kakao Mobility is foregrounding its control center.

The Open Variables

Non-technical conditions still dominate. Level 4 commercial service can only scale once regulatory approval scope, liability for accidents, and insurance frameworks are settled together. And two operational questions remain unanswered by public data: how quickly the modular kit can actually be ported across vehicle models, and whether performance holds up in daytime traffic patterns that differ sharply from the post-midnight environment being tested now.