RideFlux Turns Korea’s Driverless Permits Into A Physical-AI Test
RideFlux is pushing Level 4 autonomous-driving software from Korea’s driverless passenger permits into paid freight operations, using urban edge-case data, faster inference and licensing partnerships as its commercialization proof points.
Driverless Approval Moves From Cars To Freight
RideFlux is trying to move Korean autonomous driving beyond controlled demonstrations.
The company, led by CEO Park Jung-hee, has secured Korea's only unmanned permit in passenger transport and received the country's first approval to carry paid freight with a large self-driving truck.
That combination makes the company a test case for whether Level 4 software can operate across both passenger mobility and logistics.
The commercial claim rests on more than a route trial.
RideFlux says its technology has been trained around driverless operation in dense urban conditions, rather than simply expanding distance with a safety operator on board.
Park's argument is that mileage alone can hide the hardest part of autonomy: rare road behavior that forces the system to decide quickly without human rescue.
Jeju Data Gives The System Harder Road Cases
The company uses Jeju as a proving ground because the island offers weather and road conditions that are more difficult than a straight highway route.
RideFlux has collected cases from mountain roads, roundabouts, unsignalized intersections and construction zones, alongside urban hazards such as jaywalking pedestrians, birds entering the road and illegally parked vehicles.
That emphasis matters for Level 4 deployment because predictable route repetition can make a system look safer than it is.
RideFlux combines rule-based logic with AI judgment through a hybrid structure that selects the safer and more accurate output.
It also uses model-lightening and knowledge-quantization techniques to preserve accuracy while reducing computation demands to a one-tenth level.
The company's operating claim is specific: its urban driverless record now exceeds 3,000 hours, and it says that record has produced zero accidents.
That figure does not prove broad national readiness, but it gives regulators and fleet partners a measurable operating record under fully driverless conditions.
Inference Speed Becomes The Competitive Metric
RideFlux's clearest external benchmark came at the CVPR 2025 E2E Autonomous Driving Challenge.
The field included 29 teams, with Nvidia, Xiaomi and the Swiss Federal Institute of Technology Lausanne among the competitors, and RideFlux finished third overall.
In the inference-speed category for unexpected situations, it ranked first with 14 ms.
For autonomous driving, speed is not a cosmetic benchmark.
Park says a correct decision can still be unsafe if it arrives late, and he links delayed judgment to the sudden braking that passengers often feel in autonomous vehicles.
The company's pitch is that faster inference supports both emergency response and ride comfort, which are essential if unmanned mobility is to leave pilot zones.
Freight Could Be The First Scale Opening
RideFlux sees freight as the less crowded commercialization route.
Park says unmanned taxi services have already reached thousands of vehicles, while large autonomous trucks remain below 100 vehicles worldwide.
Park set a year-end goal of building a fleet above 20 large freight trucks, which he said would represent about 25% of that market.
That target is ambitious, but the business model is clearer than a one-off vehicle trial.
RideFlux says the same software can be adapted across vehicle types through parameter tuning, and it is widening its portfolio through ADAS software-licensing work with domestic and foreign automakers.
It also has partnerships with Kakao Mobility, Socar and Tada, giving it links to platform operators that already understand fleet deployment.
The Watchpoint Is Licensing Proof
The next test is whether RideFlux can turn permits and benchmarks into repeatable software revenue.
The company has received all-A grades in a technology assessment for a special technology listing and has attracted cumulative investment of 80 billion won, but investors and customers will still look for deployment evidence beyond the current operating record.
the larger signal is physical AI moving into regulated transport.
RideFlux's case shows how autonomy vendors are trying to prove reliability through edge-case data, inference latency and software licensing rather than headline mileage alone.
If its freight plan advances, Korea could become a more visible test market for Level 4 autonomous logistics.
