Socar, Korea's largest car-sharing company, plans to leverage 15 years of accumulated accident data to tap into the autonomous driving era. The company collects and manages data centrally from its fleet of 25,000 vehicles, a system it confidently claims rivals Tesla's capabilities, and intends to continue building on this strength. Its ultimate goal is to become Korea's leading mobility service platform, operating both car-sharing and autonomous taxi services.
According to the mobility industry on January 15, Socar held an internal presentation the previous day at its Seoul office in Seongdong-gu, unveiling the autonomous driving achievements of its Future Mobility Task Force (TF). This marks the first time Socar has publicly shared its autonomous driving strategy blueprint.
Socar launched the TF in January under CEO Park Jae-wook to identify autonomous driving as a future growth engine. The TF concluded that few companies domestically or internationally are as well-positioned for end-to-end (E2E) autonomous driving technology development as Socar. E2E autonomous driving refers to a structure where AI handles all aspects of autonomous driving, regarded as the next-generation technology succeeding rule-based models. "The model architecture for E2E has already been open-sourced through platforms like Nvidia's AlphaMayo, so now data for training models is more important than model design," said TF head Jang Hyuk. "Socar has optimal conditions for data acquisition."
Socar has strengths in three areas: large-scale vehicle operations, rich vehicle data collection, and a centralized data processing system (pipeline). "Tesla is the only company abroad that satisfies all three simultaneously, and Socar is the only one in Korea," Jang emphasized. Autonomous driving technology firms are limited in fleet scale, taxi platforms in the variety and quality of collectible data, and automakers in centralized data pipelines.
In contrast, Socar directly owns 25,000 vehicles. Each vehicle generates over 100 types of data including speed, steering, and braking. This is possible because the company attaches telematics terminals and front and rear dashcams for remote control and management—a structure enabled by being a car-sharing company. The TF has also built a pipeline to process this raw data into formats suitable for E2E model training.
The accumulated accident data also differentiates Socar. To advance E2E AI models, companies need as many "edge cases" as possible—data capturing exceptional situations such as pedestrians or other vehicles cutting in. Socar has collected 220,000 edge cases through insurance processing and other means following vehicle accidents. To further improve data quality, the company plans to gradually deploy 1,000 "full sensor kit vehicles" equipped with LiDAR and seven cameras, making them available to users starting this year.
Socar's ultimate goal is to become a mobility platform offering both autonomous taxi (ride-hailing) and car-sharing services. "We are considering all possibilities including partnerships with automakers and technology companies," said Lee Jeong-haeng, head of technology at the Future Mobility TF. "We will announce specific partnership and service plans as they are finalized."
Having experienced the painful shutdown of its "Tada" service in 2020 due to backlash from the taxi industry, Socar plans to be more attentive to harmonizing with existing industries this time. Lee, who led the development of Tada in the past, said, "We are communicating with stakeholders who are voicing various concerns. Given that social consensus is needed, we will actively work to resolve issues."
