Ilsu Park runs a company that puts self-driving trucks on public highways and gets paid for the loads they carry. That sentence is doing a lot of work, because most of autonomous driving still lives in demos and press releases. Mars Auto, the Seoul startup Park co-founded and leads, has moved past that stage. Its trucks run commercial freight in Korea and, since 2024, in the United States.
The core idea is deliberately stubborn. Where much of the industry builds around LiDAR sensors and centimeter-accurate HD maps, Park's team uses cameras and an end-to-end AI model that takes in what the truck sees and decides how to drive. No expensive laser rigs. No pre-mapped routes to memorize. Park likes to compare the approach to Tesla's: vision in, decisions out. The payoff he points to is blunt - automating a Class 8 truck for roughly $7,000 a unit, a number that makes the math of a trucking business look very different.
That choice is the whole strategy. Cheaper hardware means a lower bar to put a truck into revenue service, and it means Mars Auto can run those trucks itself rather than waiting to sell software to someone else. Park is explicit that he is not in the business of shipping autonomy kits. He is in the business of moving freight, cheaper.
The distinction matters more than it sounds. A company that sells autonomy hardware depends on convincing fleet owners to gamble on unproven technology, and its revenue arrives only after a long sales cycle. A company that owns the trucks and sells the freight service captures the savings directly and controls the whole operation - route planning, dispatch, the truck itself. When the automation improves, the margin improves with it. Park has built the business so that every gain in the software shows up as a lower cost per mile rather than a better spec sheet for a customer to evaluate.
From a video game to real 18-wheelers
Park's path into the field started without a truck at all. Around 2016, before Mars Auto existed, he built an open-source project called europilot - a toolkit that trains an end-to-end driving model inside Euro Truck Simulator 2, the trucking video game. It let anyone teach a neural network to steer a simulated rig. The project still carries a truck emoji as its signature and has collected roughly 1,500 stars on GitHub, where Park goes by the handle daftshady.
The through-line from that hobby to the company is unusually clean. The simulator taught a model to drive from pixels; Mars Auto does the same thing with real cameras on real trucks. Park studied computer science at KAIST, and it was there he met co-founder Gyuri Im, who shares the same academic background. In 2017 they turned the idea into a business.
Park's engineering roots run through everything the company does. His public code goes back years - the truck-simulator toolkit, a small async web framework, the usual pile of configuration files that mark someone who lives in a terminal. He is a builder first and a chief executive second, and Mars Auto reflects a founder who would rather ship a working model than commission a study. Choosing cameras over lasers was, at bottom, an engineer's bet that a good enough model fed enough data will beat a more expensive sensor stack that still needs a model to interpret it.
Building in Seoul, running in Texas
Mars Auto went through Y Combinator in 2019 and drew backing from investors including Kakao Ventures, the venture arm of the Korean tech giant. Total funding sits at around $12.5 million, including a Series A in 2022. In 2021, Park landed on the Forbes 30 Under 30 Asia list in the industry and manufacturing category.
The operational milestones are what set the company apart from pure research shops. Mars Auto became the first in Korea approved to run Level 3 self-driving trucks on highways, with a safety driver aboard to take over if needed. It began commercial autonomous freight in Korea in 2023 under a regulatory sandbox. Then it did the counterintuitive thing: it opened paid US operations in 2024, basing itself in Round Rock, Texas, and targeting the Texas Triangle freight corridor that links Dallas, Houston, and San Antonio.
Highways are the point. Long-haul routes are the most repetitive, most standardized part of freight, which makes them the easiest to automate and the most expensive to keep staffing. Park frames the industry's chronic driver shortage not as a crisis to lament but as a market waiting for a cheaper answer. The company leaves the hard, unstructured last mile to humans and goes after the 99% in between.
The move into the United States raised eyebrows for a reason. American autonomous trucking is a crowded, heavily capitalized field, and plenty of local players with far bigger war chests have stumbled or shut down. For a Korean startup with about $12.5 million raised to plant a base in Texas and start hauling paid loads is either reckless or a signal of real confidence in the cost structure. Park's argument is that the cheaper the automation, the less capital it takes to reach break-even on a route - and the Texas Triangle, with its dense, high-volume corridors between major cities, is a good place to prove that arithmetic.
The data, and the deadline
Behind the pitch is a growing pile of kilometers. Mars Auto says it has accumulated about 20 million kilometers of real truck-driving data and 2.5 million kilometers driven autonomously - the kind of real-world exposure that a camera-and-AI approach needs to keep improving. That data moat is the argument for why Park believes full autonomy is reachable rather than perpetually five years away.
His deadline is specific: fully driverless long-haul trucks by 2028. In 2026 the company cleared a mobility demonstration exemption and passed an evaluation by the Korea Transportation Safety Authority, and it lined up three clients to launch what it describes as Korea's first paid autonomous trailer transport, running to and from Busan Port. Each of those steps is a rung on the ladder from Level 3, where a human still sits ready, to a truck that drives the highway alone.
The Busan Port work is worth pausing on, because ports are where freight economics get real. Container terminals run around the clock on predictable routes with heavy, repetitive traffic - the exact conditions where a self-driving trailer earns its keep. Landing three paying clients there, rather than staging another demonstration, is the kind of proof point Park favors. It also fits a pattern in how he talks about the company: not as a moonshot chasing a distant breakthrough, but as an operator stacking concrete regulatory approvals, real contracts, and logged kilometers until the driverless step becomes the obvious next one rather than a leap of faith.
Park's ambition reaches past any single corridor. The long version of the plan is freight that moves itself between Korea and the United States, with software good enough to run either country's highways and a cost structure low enough to make the service cheaper than the human alternative on both. That is a large claim from a small team, and Park has never pretended otherwise. What he offers instead of hype is a trail of milestones that keep arriving on schedule.
There is a real technical argument underneath all of this, and it is not settled. The camera-first camp holds that human drivers manage with two eyes and a brain, so a system with more cameras and a strong model should eventually match and exceed them - and that piling on LiDAR and HD maps buys expensive crutches that break the economics. The other camp counts on those sensors for the redundancy that safety cases demand, especially in bad weather and low light. Park has placed his company firmly on the cheaper, harder side of that line, which means the burden of proof sits with him. The 20 million kilometers of collected data is his response: exposure to enough real situations that the model learns the long tail rather than being told about it.
Whether the camera-only bet fully pays off is still an open question, and the trucking-autonomy field is littered with well-funded companies that promised more than they shipped. What separates Park is that his trucks are already logging paid miles while the debate continues. He built a freight service first and is letting the autonomy catch up to it, one route at a time.
If he hits the 2028 goal, Mars Auto becomes one of the few companies anywhere running driverless heavy trucks as a paying business, and it does so from an unlikely origin - a Seoul startup that got its start in a video game. If he misses, the miles still accrue and the freight still moves with a driver aboard, which is a softer landing than most autonomy bets allow. Either way, Park has spent nearly a decade turning a stubborn idea into trucks on the road, and he is not slowing down to argue about sensors.