Hanbin Lee, CEO: “We are in the final contract stage with automakers for autonomous driving within factories” [WE, are not ROBOTaxi ]

Tesla’s robotaxi has often been compared to ChatGPT in the realm of autonomous driving. While traditional technologies learned road rules and situations step by step like humans, Tesla’s technology directly learned human driving habits. As a result, while conventional robotaxis needed to adapt anew to each country or city, Tesla’s robotaxis did not.

Tesla’s robotaxi, set to debut in Hollywood on October 10, was highly anticipated for these reasons.

During the event, Musk promised a 2026 release but provided no detailed plans. Market expectations of a partnership with Uber were unmet. The showcased robotaxi operated solely on pre-designated paths within a set.

Entrepreneurs in Korea’s autonomous driving technology industry shrugged off the announcement. “The full-scale commercialization of Tesla’s robotaxis is still years away,” said Hanbin Lee, CEO of Seoul Robotics.

They don’t claim to overtake Tesla. Rather, they are confident in building businesses faster than big tech by leveraging their own comparative advantages.

Just as reaching a destination doesn’t always require highways, the path to full autonomy doesn’t necessarily have to involve robotaxis.

● Hanbin Lee, CEO of Seoul Robotics Born in 1991. Studied abroad in Boston during middle school. Majored in mechanical engineering at Penn State University and founded the company in 2017. That same year, it ranked first in the Lidar-based autonomous driving category at the “Self-Driving Car Challenge” in Silicon Valley.

At Seoul Robotics’ test track in Gwacheon, Gyeonggi Province. In the 4,500㎡ experimental area, five cars were slowly moving in a queue. All five cars were unmanned and autonomously driving.

However, there were no sensors or Lidar typically associated with autonomous vehicles. Lidar is usually visible as horn-like structures on the roof or outside the rearview mirrors, making autonomous vehicles conspicuous. The cars in this area weren’t hiding their sensors; they were standard models purchased directly from dealerships.

Hanbin Lee, CEO of Seoul Robotics, explained, “It’s like remotely controlling vehicles as if they were RC cars.”

The remote control isn’t done by humans but by the “infrastructure.” A central control system uniformly operates the vehicles. Sensors installed at each corner of the test site transmit data to the system, which interprets the environment and moves the vehicles accordingly. While it is indeed unmanned autonomous driving, it leans more toward “infrastructure driving” than “self-driving.”

CEO Lee emphasizes the significant utility of infrastructure-based autonomous driving (hereafter referred to as "infrastructure driving"). This technology enables vehicles to move from production lines to storage yards and onto export ships without human intervention. Since no vehicle modifications are required and the central system can control multiple vehicles simultaneously, it streamlines the process. He hinted, "We are conducting pilot tests (Pilot 2) with several global automakers."

In Pilot 2, the technology is tested in actual factory environments. Typically, technology adoption follows a progression from proof of concept (PoC) through Pilot 1 and Pilot 2.

“We’re at a stage where we can visualize the size of contracts in Europe, Japan, Korea, and the U.S.,” said Lee, who plans to go public on KOSDAQ next year. A key focus is ensuring that contracts are publicized before the listing. He expects results “within a year.”

Skepticism exists among investors. This is because Seoul Robotics’ infrastructure driving technology and its target market differ from the autonomous driving trends represented by Tesla and Google’s Waymo. These big tech companies focus on enhancing the autonomous driving capabilities of individual vehicles and aim for public road autonomy rather than limited locations. Beneath their concerns lies the recurring question: “Can a Korean company compete with big tech in terms of technology?”

For eight years since its founding, Lee has faced numerous challenges and concerns about his company and Korea’s autonomous driving industry. He asserted, “We’ve firmly chosen an alley worth fighting in.”

Q: Is it a niche market?

Compared to robotaxis, the market may be smaller, but the demand is strong. It’s a “4AM CLAIM” market, meaning it’s an urgent problem worth a call even at 4 a.m.

“The story will be, ‘Seoul Robotics was the first to commercialize infrastructure driving technology in this market, and multiple global automotive OEMs are working with us.’ That will become the brand of Seoul Robotics. For buyers, trustworthiness is as important as technology. That’s the essence of brand power. We’ve chosen the alley where we can fight. Here, we’ll build a solid brand.”

Once the brand is established, our technology can be applied to logistics centers, ports, and airports.

Q: What problems do automotive companies want to solve?

Currently, human drivers move cars from assembly lines to export ships. Factories employ hundreds of drivers for these tasks. Labor costs are an issue, but finding people to do the work is even harder. German factories, for instance, bus in workers from Eastern Europe every morning. Logistics centers face similar challenges, with one in three workers quitting within three months. This isn’t about eliminating jobs but providing solutions where labor shortages exist.

Q: Machines might be more expensive than humans, though.

To illustrate, in moving vehicles 50 meters from factories to storage yards, 12 drivers usually work in three shifts. Operating this over a year costs several hundred million won. Applying Seoul Robotics’ infrastructure driving can cut the actual operating costs by more than half. We install sensors, fiber-optic networks, and control software. From the first year of installation, cost savings are achievable.

Q: I noticed the technology is also applied to production lines in the diagram. Aren’t these processes already automated with conveyor belts?

Electric vehicles can move with just a motor and a battery. By enabling EV platforms to autonomously navigate assembly lines, frames can be mounted, and doors attached as needed. This replaces conveyor belts. Tesla is also experimenting with this concept, called "SDM (Software Defined Manufacturing)." Tesla unified the hardware systems inside its vehicles to make the car operate like a single computer, allowing software to easily control it—this was SDV (Software Defined Vehicle). SDM is the factory version of this concept.

Seoul Robotics is confident it can help OEMs implement SDM by integrating fragmented systems within factories and operating them through a central operating system (OS) for management.

Q: Global parts companies like Bosch and Continental have entered the competition. Can you maintain a competitive edge?

Our advantage lies in the data we’ve accumulated on-site. Using this data, we’ve improved the technology, lowering costs and enhancing performance. For instance, Seoul Robotics installs one sensor every 30 meters, whereas competitors need at least 12 sensors to achieve similar functionality. In autonomous driving, the first step—perception—isn’t just about installing more sensors. The software must be capable of learning and improving using data. For example, if sunlight reflected off a building’s glass creates sensor errors, the software needs to understand and handle such situations.

Because we’ve enhanced software performance with accumulated data, unlike others, we can perform infrastructure-based autonomous driving outdoors (Level 5 Control Tower). This allows autonomous driving from factories to storage yards.

Finally, we’ve implemented "fleet control," enabling the central control system to operate multiple vehicles simultaneously. In theory, hundreds of vehicles can be managed at once. While other companies require humans to remotely control their vehicles, ours operate without humans inside or outside the vehicles.

A Seoul Robotics external board member, who has previously worked on wireless communication technology for global tech firms, highlighted the fleet control technology and its market potential.

Having first met CEO Hanbin Lee at a Korean startup event in Silicon Valley last year, he stated, "Communication between infrastructure and vehicles is already a universal technology," but added, "Connecting, controlling, and managing hundreds or thousands of vehicles simultaneously in a confined space requires a highly advanced level of technology."

He further explained, "Infrastructure-based driving is a high-difficulty service model that requires both software and hardware expertise." Given the variety in vehicle types and models, communication systems differ in performance and functionality. To simultaneously control such vehicles, "You need to understand the communication systems of all the vehicles, establish a wireless network considering radio interference, and, most importantly, provide competitive wireless and sensor infrastructure at a reasonable cost," he said. He concluded, "For manufacturers, collaborating with Seoul Robotics, which has secured ultra-gap technology in the B2B autonomous driving market, is inevitable."

Q: You’re in a profit-driven business, yet you’re not fully earning revenue yet.

We’re right on the verge. The technology has been verified. Outdoor driving and fleet control have been proven. We’re currently reviewing final contracts. We’re seeing the scale of these contracts firsthand in Europe, Japan, Korea, and the U.S. Each contract is worth at least hundreds of billions of won. Moreover, in this market, we have the most customer references.

Q: What aspects are currently being negotiated?

The allocation of liability in case of accidents and the unit cost per vehicle are being negotiated. Discussing unit costs per vehicle during a project is a strong indicator that the contract is nearing completion.

Initially, Seoul Robotics also envisioned autonomous driving on public roads. After its founding in 2017, the company carried out around 100 projects over four years. However, CEO Lee remarked, “The market wasn’t moving.”

“For automakers, autonomous driving on public roads was a ‘nice-to-have’ but not an urgent technology. Most companies didn’t invest in such problems,” he said.

Despite the lukewarm market, Google’s autonomous vehicle subsidiary Waymo has persisted. Starting development in 2009, it launched pilot operations in 2021. This year, as full-scale operations began, Waymo receives 100,000 ride requests weekly. Google has invested approximately $4.75 billion to date and plans to invest an additional $5 billion.

“They had more data and invested billions, while our team was fewer than 50 people,” Lee said, acknowledging the competition's scale.

When asked whether he considered this a failure, he said, “I think one of my theories turned out to be wrong,” adding, “Making a living from autonomous driving software on public roads is not our focus for now.”

As concerns grew, an opportunity came from BMW, which sought a company to develop infrastructure-based autonomous driving services within factories. Korea’s KIAT (Korea Institute for Advancement of Technology) facilitated the partnership in 2020 through an international joint technology development program connecting global value chains with skilled Korean SMEs.

Thanks to this breakthrough, Seoul Robotics is now in the final stages of testing driving performance inside actual factories with multiple OEMs.

“To solve (autonomous driving) problems in factories, I realized that an infrastructure-based approach was better suited than a vehicle-based one,” Lee explained.

Q: Waymo has successfully commercialized its technology. If vehicle-based autonomous driving improves further, will external infrastructure still be necessary?

Applying Waymo’s system to B2B environments like factories or logistics centers would be more expensive, slower, and less reliable. It’s similar to the relationship between Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs). Both are transport robots, but AGVs follow pre-laid magnetic tracks, while AMRs navigate independently without tracks. Our technology is akin to AGVs, while Waymo resembles AMRs.

Though AMRs are smarter, they’re not always preferred in industrial environments, which value safety and speed over sophistication. AMRs often operate slower than their potential speeds due to frequent errors. When AMRs emerged in the late 1990s, it was expected they’d replace AGVs, but that prediction fell short. This trend will likely continue in autonomous driving.

The future may not converge into a single system but rather combine the strengths of AGVs and AMRs. For example, AGVs could be equipped with high-performance sensors.

Q: But if Elon Musk delivers on his promise to offer robotaxis under $30,000 (with a fare of $0.20 per mile), wouldn’t that change things? Errors will also decrease over time.

In our system, 90% of CPU usage is consumed by perception software. However, whether we control one or ten vehicles, the CPU and memory usage doesn’t significantly change because we operate on an infrastructure-based model. In contrast, vehicle-based autonomous driving requires equipping each vehicle with autonomous driving capabilities, which increases costs as scale grows.

Q: Instead of bringing Waymo into factories, could infrastructure-based driving be applied at a city level?

Factories span about 1 square kilometer, which is essentially a small town. Since we’ve gained experience, scaling up to 2 or 5 square kilometers isn’t a major challenge. For example, infrastructure-based autonomous driving could be implemented in speed-restricted zones like school zones or residential complexes. In areas with speed limits under 30 km/h, such systems could control vehicles.

However, I don’t believe infrastructure-based autonomous driving alone can cover vast areas in the long run. Just as AGVs and AMRs coexist in factories, I foresee a middle ground on public roads. The structure will likely involve collaboration between vehicle-based autonomous driving and infrastructure-based systems.

Lee identified the domain of infrastructure-based driving. He also emphasizes the scalability of B2B autonomous driving, expanding beyond vehicle delivery to valet parking in parking lots and truck driving within logistics centers.

However, some investors still worry about the company's valuation—not the market, but the technology. They believe technological trends are shifting toward "machine learning-based autonomous driving."

Traditional autonomous driving technologies, including Waymo's, follow three sequential steps: perception → decision-making → control. Sensors collect environmental data, and perception software processes the noise. Decision-making software compares this processed data with pre-stored 3D maps and determines how to act (e.g., accelerate, brake, or steer) based on pre-programmed road rules. Machine learning techniques are used, but only to improve the performance of individual software components.

In contrast, Tesla's machine learning-based autonomous driving combines perception and decision-making into one. It learns by observing human driving habits in environments detected by sensors. Consequently, the system can act like a human once the sensors collect data. Efforts such as coding condition-specific behaviors and pre-building 3D maps are replaced by neural networks. Lee compares this to multimodal models in generative AI.

The "tech winter" remains ongoing, but machine learning-based autonomous driving has been an exception. The UK-based company Wayve, considered a frontrunner alongside Tesla in this field, raised $1.05 billion in a Series C funding round last May (approximately 1.438 trillion won). Investors included SoftBank, NVIDIA, and Microsoft. Hyundai Motor Company is also reportedly considering investing.

Wayve refers to its technology as "end-to-end AI," meaning it processes everything from data collection to decision-making on-site.

Even before entering the infrastructure-based driving field, Seoul Robotics was regarded as a company with excellent perception software technology. But now, as AI takes over both perception and decision-making, has the company's strength become an outdated trend? Lee argues, "90% of autonomous driving problems are still perception-related."

Q: What does that mean?

Think of the game StarCraft. When you designate a destination for a unit, it autonomously reaches that point. It works because the game already knows the locations of in-game obstacles. In other words, once perception issues are resolved, the next steps are not difficult.

But reality is not as simple as a game. Unexpected situations occur, and data can become noisy. How would an autonomous vehicle respond if an elephant suddenly appeared in the middle of downtown Seoul? It would struggle. Regardless of the autonomous driving method employed, such situations must be anticipated, and the software must be further refined. For example, Tesla’s AI team focuses 90% of its efforts on perception issues.

Q: Couldn’t performance be improved by attaching more sensors, even if it’s costly in the short term?

That’s a Korean mindset—focusing on hardware development. The reason Korean autonomous driving companies struggle despite pouring in massive capital is because of that approach. You need to build solid software starting from the perception stage.

Q: It seems like it will take time to solve the problem.

It reminds me of a gradually flattening upward curve. If 100 is the goal, reaching 90 is relatively easy, but increasing the number beyond that becomes difficult. It will take time for them to solve perception issues and realize autonomous driving services on public roads. I see an opportunity for us at that point—commercializing quickly, building a brand, and expanding our domain.

Technology is a tool for proving a hypothesis. The hypothesis I ultimately want to prove is this: “A global tech startup can emerge from Korea.”

Q: You emphasize branding. Is a "B2B autonomous driving brand" your ultimate goal?

No. Technology is a tool for proving a hypothesis. The hypothesis I ultimately want to prove is this: “A global tech startup can emerge from Korea.” Korea’s Samsung Electronics and Hyundai Motors built global businesses based on the domestic market. But it will be difficult moving forward because the domestic market is shrinking.

Q: It reminds me of the first-generation entrepreneurs’ phrase, “serving the nation through business.”

I like the books written by the late Chairman Ju-yung Chung. Thanks to entrepreneurs of that era, Korea has come this far. However, I believe Korea has already passed its golden age. The peak was in 2020. Like Europe, it might either decline or maintain the status quo. To maintain the status quo, someone needs to have a clear vision for the future and create a new model of success.

Q: Despite the philosophy, your outfit is quite unique. You insist on wearing a Boston Red Sox cap and Crocs shoes.

The cap and Crocs are part of my past and are also a way to represent Seoul Robotics.

When I was in middle and high school, I used to watch Boston Red Sox baseball games once or twice a month. I would go to the stadium and sing "Sweet Caroline" (the Red Sox anthem). That became part of my identity. And when I went to engineering school, everyone was wearing Crocs.

I used this as part of the company’s identity. I believe a consistent image is necessary for a brand to be recognized and remembered. With great products to back it up, it could become a strong brand.

However, tech companies didn’t seem to care much about branding. ZEISS lenses and lenses made by domestic companies are technically quite similar. The price difference, however, is significant. In the end, tech companies also need a brand, and to build a brand, you need a "Mario." There must be a face for the company. Since Americans have a hard time distinguishing East Asians, I used this point. Just like Mario’s symbols are his mustache and cap.

Q: You don’t take it off even when you go to the presidential office?

When people feel that "he will take it off when asked," the authenticity of the brand disappears.

Q: You’re preparing for a technology special listing. If you generate revenue soon, there might be other options.

In the domestic market, I think the technology special listing is essentially gone. They evaluate based more on revenue than technology. And realistically, the upper limit of corporate value is considered to be 300 billion won. But Seoul Robotics has already been recognized at that level. From an investor's perspective, they are concerned about whether they can make returns higher than their investment. The company should be valued in the trillions of won.

So, why insist on a technology special listing? I don't think it’s necessary to avoid it just because of bad precedents. Honestly, there are times when I think a NASDAQ listing might be easier. Foreigners are pulling their money out of Korea. But on the other hand, with both technology and revenue, we could break through that barrier. I don’t think the easy path is always the best.

The content above is a translation of the original source. To view the original content, please click the links below.

https://www.fortunekorea.co.kr/news/articleView.html?idxno=43681

https://www.fortunekorea.co.kr/news/articleView.html?idxno=43684