A Day in Seoul’s First Robotaxi
<Launching robotaxi>
On September 26th, Seoul launched its first robotaxi service in a geofenced area covering about 9 square kilometers, which includes the Gangnam Station, one of the busiest places in Seoul. I was eager to take a ride and see how self-driving technology is progressing in South Korea.
<Going for a ride>
After work, I headed to Gangnam and met my friend L, who works nearby. We initially planned to eat spicy pork bone soup, but the restaurant had a long queue, so we opted for a place that served pork soup instead. The meal was delicious, and I particularly enjoyed the spicy beef soup noodles, which were better than my dish. We caught up over dinner and then moved to a cafe owned by my friend’s company. Enjoyed hot Americanos. Afterward, my friend went home, and I waited for the robotaxi.
<Robotaxi operating time>
The robotaxi service operates only late at night on weekdays, from 11:00 PM to 5:00 AM, so I had to wait until then. At 11:00 PM, I tried to hail the robotaxi using the app called Kakao T, but it showed no cars available. With only three robotaxis in operation, it was challenging to catch one, especially since it was free and many people were trying to use it due to the high cost of taxis late at night in Seoul.
<Eager to find robotaxi>
I decided to move one station further from Nonhyeon Station to Gangnam Station, but I still couldn't get a robotaxi. I then realized that the answer was to move from a busy area to a residential one, as most people were heading home at that time. So, I took the subway to a residential area. Unfortunately, I had to transfer subway, and there were no other subways available when I arrived at the transfer point. I stayed on the subway until I reached Samsung Station, located at the geofence border. At Samsung Station, I still couldn't find a robotaxi, so I took a bus to Daemosan Entrance Station, located in the far right and bottom of the geofenced area. I figured this would be a less competitive spot to hail a robotaxi, as people would be dropping off rather than hailing one in this residential area. Finally, at 12:15 AM, I succeeded in hailing a robotaxi and was thrilled to take a ride.
<Riding a Robotaxi>
The robotaxi arrived about five minutes later. I got permission to take the video of the interior and the driving experience from inside the car. The ride took about 18 minutes to reach my destination. According to the operator, most people take the robotaxi near Gangnam or Yeoksam Station. Tip for getting a robotaxi easily would be to avoid hailing it at Gangnam or Yeoksam Station. Residential areas at the bottom of the geofenced area are less competitive.
The robotaxi is developed by a self-driving tech company called SWM. A newcomer from SWM was seated in the co-pilot seat, while the driver, who was the PM of the company robotaxi project, kindly answered all my questions.
SWM's self-driving R&D center employs around 90 people, working on hardware and software solutions. They started developing this technology in 2017, creating geofenced areas and high-fidelity maps. The self-driving system operates with just a start and end point, and the navigation route is produced immediately after input. They use market-available sensors, but the main computing program, including voltage and network management, is self-developed, minimizing voltage and network issues. The system can predict nearby vehicle behavior up to 7 seconds ahead and make self-lane changes. Their software and sensors are compatible with various vehicle types, and they can even drive through tunnels without GPS.
<Start trip>
After fastening my seatbelt, the driver enabled the self-driving mode with the press of a button.
<Vehicle Info>
The vehicle, a SsangYong Korando EV, is a small electric SUV equipped with 8 lidars (4 long-range and 4 short-range) and 10 cameras. It has a total of 30 sensors, including lidars, cameras, radar, and ultrasonic sensors, providing 360-degree coverage around the car.
<False Lane-chaning>
The robotaxi successfully performed all lane changes, though they weren’t smooth due to the large steering angle.
<School Zone Manual Driving>
When entering a school zone, the system automatically turned off the self-driving mode, as South Korean law requires manual driving in school zones with a 30 km/h speed limit.
<Intervention options>
For interventions, the operator could use the steering wheel, brake, accelerator, or the emergency button on the UI.
<UI, Traffic Light>
The UI also displayed the vehicle's speed, steering angle, and pedal pressure. While the layout resembled Tesla's interface at a glance, it had a key difference, the ability to detect traffic lights in detail, including left-turn signals and the various colors of lights. However, it currently cannot detect pedestrian traffic lights. They use both cameras and V2X infrastructure, which connects with traffic lights to know when they will change. It performed well in the dilemma zone at intersections with yellow lights.
<AI-based software>
The self-driving software is AI-based, trained with data collected from their three vehicles. They update the software every two weeks.
<False Traffic enforcement Camera>
However, it doesn't recognize traffic enforcement cameras and always drives under the speed limit, which might be a drawback in the future, as excessively slow driving can disrupt traffic flow.
<False Road Signs>
In the future, they plan to detect visual speed limits and road signs. The vehicle can make unprotected left turns, U-turns, and right turns, including right-on-red.
<False Construction Site>
Near the end of my journey, a critical failure occurred. There was a guard rail with a construction site along the right lane, and the car suddenly steered towards the guard rail. I think the car was trying to change lanes to the right before making a right turn but failed to recognize the guard rail. The operator intervened immediately to avoid a collision.
<Summary>
Overall, it was a great experience riding the commercially available robotaxi in Seoul, even though it’s still in a pilot phase. The strengths were its accurate traffic light recognition using V2X infrastructure and its focus on safety, despite being slow. However, it still struggles with recognizing construction sites, curbs, and non-vehicle objects, indicating that collision avoidance is too far from perfect. The road environment is always variable, with many situations involving people and objects beyond just vehicles. For true, safe autonomous driving, more data is needed, and with only three vehicles, large-scale data collection remains a big huge challenge. But it’s a significant step forward for South Korea’s autonomous driving industry. It may be late, but better late than never. We can do it.
