South Korea's mobility platform industry is expanding its autonomous driving business by using proven data as a weapon. As the auto industry is reshaping from hardware to software and services, a broader push is spreading across the sector to turn real-world driving data accumulated through in-house services such as car sharing, ride-hailing and shuttles into artificial intelligence training assets.

According to the industry on May 15, Socar is setting up an independent corporate entity for autonomous driving, and Kakao Mobility has formalised a shift into a physical AI company. Autonomous driving software company Autonomous A2Z is also moving to build a mass-production system by teaming up in succession with automakers and parts makers.

Socar will set up an independent company dedicated to autonomous driving services in May. The investment will total 150 billion won, which the company described as among the largest in South Korea's autonomous driving services sector. Socar CEO Jae-wook Park (박재욱), who has overseen the new autonomous driving business since January this year, will concurrently serve as head of the new entity.

Krafton will join the new company as a strategic investor. It will participate in Socar through a 65 billion won third-party allotment paid-in capital increase and also make a separate investment in the new entity. Krafton judged it needed a partner with operating capabilities in the offline, physical domain to commercialise autonomous driving services. It chose Socar because Socar has experience managing large numbers of vehicles and drivers through operations such as Tada in the past. Krafton plans to use real-world driving data accumulated during the operation of the joint venture in its research on physical AI, or artificial intelligence based on the physical world.

The competitiveness of the new entity lies in Socar's data assets built over 15 years. Socar said its Future Mobility task force in the first quarter built a centralised data pipeline that collects in real time about 1.1 million km of real-world driving data a day based on a car-sharing fleet of 25,000 vehicles. It also secured edge-case data such as 220,000 cases of accident data. The new entity will expand its business scope step by step, starting with Level 2 car-sharing services and moving to Level 4 fully autonomous ride-hailing. Socar on May 6 invited executives of the Korea Rent-a-Car Cooperative Federation and held a demonstration ride event on an approximately 4.5-km route around the Hwaseong City autonomous driving living lab area using a Torres EVX-based autonomous vehicle.

Kakao Mobility has declared a shift into a physical AI company. The company held an all-hands meeting for all employees last month at its Pangyo office, where Vice President Jin-gyu Kim (김진규), who also heads the physical AI division, shared the autonomous driving strategy and technology direction in person. "Based on the Kakao T platform's data and technology capabilities, we will add a new technological value called physical AI to drive new mobility innovation," Kim said.

Its technology advancement plan has three pillars. It will combine its own autonomous driving technology with the Kakao T platform infrastructure to build a large-scale data pipeline and advance the core end-to-end autonomous driving model. The targeted range of technologies covers all areas from software to hardware. Kim said it will upgrade the planner, a key factor in autonomous driving vehicle decision-making, with high-quality data and apply it sequentially to services in the Gangnam area. It will also strengthen external cooperation. It plans to build an open ecosystem by expanding domestic autonomous driving partnerships that have continued since 2020.

It is also moving actively in fields of use. Autonomous driving software specialist Autonomous A2Z, or A2Z, is focusing on securing mass-production systems and control technologies. A2Z signed a memorandum of understanding with KG Mobility and KGM Commercial to develop Level 4 autonomous vehicles and key parts. It will combine A2Z's full-stack autonomous driving technology with KGM's vehicle design and production capabilities and KGMC's commercial vehicle and electric bus technology to build a Level 4 mass-production system based on electric buses. The scope of cooperation spans the full cycle, from joint parts development to securing performance certification and creating follow-on businesses.

A2Z has also teamed up in succession with HL Klemove and HL Mando. With HL Klemove, it will jointly develop an end-to-end autonomous driving system by combining radar, camera, high-performance controller and sensor fusion technologies with data from the autonomous shuttle ROii that is operating in demonstration runs at Cheonggyecheon. With HL Mando, it will work to secure stability in the control domain, the final stage of the perception-judgement-control mechanism, based on an electronic power steering system. "Autonomous driving is a representative technology convergence industry in which close integration between software and vehicle parts is key," A2Z CEO Ji-hyung Han (한지형) said, adding, "We will raise the actual service applicability of Level 4 autonomous vehicles."

◆Why autonomous driving is accelerating: restructuring of the auto industry's profit model is the backdrop

Behind the mobility industry's accelerated push into autonomous driving is a shift in the auto industry's profit structure. According to Eugene Investment & Securities, the revenue mix of automakers is expected to shift from 75 percent new car sales (hardware), 15 percent after-sales service and internal combustion engine parts, and 10 percent software and finance to 45 percent new car sales, 35 percent software subscriptions and AI services, 15 percent data and MaaS, and 5 percent after-sales service and others. The trend is moving from a one-off structure in which most revenue occurs at the point of vehicle sales to a circular profit model in which revenue occurs throughout the vehicle life cycle. Another factor accelerating the shift to software is that hardware manufacturing margins are around 5 to 10 percent, while operating margins for AI software subscription services exceed 40 percent.

The common point among the three companies is that they use real-world driving data accumulated from operating their own services as a key asset for AI training and validation. The nature of the data varies. Socar has everyday driving and accident data based on its car-sharing fleet, Kakao Mobility has ride-hailing-based urban transport data, and A2Z has shuttle data from Cheonggyecheon and data from urban demonstration vehicles. As users can perceive large performance differences even for autonomous driving technologies at the same level, the volume and quality of proven data are emerging as variables that determine success or failure of commercialisation.

The establishment of mass-production systems, data standardisation and regulatory improvements are expected to influence the timing of autonomous driving commercialisation going forward. "Because users of an autonomous shared-car service can act as both passengers and drivers, unlike robotaxis, it can be applied even in the early stages of Level 4," Kang- pyo Kang (강경표), a team leader at the Korea Transport Institute, said.