Mars Auto, a South Korean startup building autonomous-driving software for heavy trucks, says it will release what it calls Tesla-comparable self-driving software this year and aims to automate freight all the way from Korea to the United States. At a media day in Seoul, CEO Park Il-su said the company will commercialize driverless trucks by 2028 using an end-to-end (E2E), camera-based AI system he likened to Tesla's, according to Seoul Economic Daily — processing visual data through cameras and letting AI make decisions, rather than relying on LiDAR and high-definition maps. Founded in 2017 by KAIST alumni, Mars Auto has real operations to point to, though its most eye-catching framing is its own. Why Cameras, and Why Trucks The company's bet rests on two technical arguments. The first is architectural. Most autonomous systems pair LiDAR — laser sensors that build a precise 3D picture — with high-definition maps, and can operate only on routes prepared in advance. Mars Auto instead uses a single neural network that takes raw video from the truck's cameras and directly outputs perception, judgment, and control, with no pre-built map, which it says lets the truck drive roads it has never seen, the way a human does. Mars Auto argues the camera approach wins on both capability and cost: where rivals' sensor setups can run around 350 million won ($229,000) per vehicle, it says its own runs about 10 million won ($6,500). The trade-off, and the reason this approach is debated, is that cameras alone can struggle in fog, glare, or darkness where LiDAR's active sensing still works — the same camera-only philosophy Tesla uses is currently the subject of a U.S. federal safety investigation. The second argument is about problem choice. Because roughly 98% of a truck's route is highway — a far more uniform