Building autonomous driving software that holds up outside the lab.
I'm a 3rd-year Automotive Engineering student at Kookmin University. I develop perception and control systems for scale cars through KUUVe, the autonomous driving research club I lead. I care about the gap between benchmark numbers and real-world field behavior, and about building the process that closes it.
Tracing failures on a real car is what led me to software verification: designing tests from requirements, injecting faults on purpose, and measuring whether the tests themselves are any good.
Autonomous driving stack — Integrating perception, localization, and control end-to-end on a real vehicle and validating behavior through repeated field tests.
Edge AI deployment — Getting inference models onto constrained hardware and verifying that speed and accuracy hold across devices and conditions.
Field fault isolation — Tracing failures from symptom to root cause through logs, layer by layer, until the fix is provable and the regression is documented.
Three projects verify the same automatic emergency braking function at three levels: the unit, the ECU on its network, and the test design process itself. A fourth takes the question to real hardware: do the safety mechanisms of an ECU work when the fault actually happens?
| Project | Level | What it does |
|---|---|---|
| automotive-sw-qa | Software unit | Requirement-based testing of one function in three forms (Simulink/Stateflow model, C code, Python reference) compared back to back, with the test design measured by coverage up to MC/DC and by mutation testing. |
| ecu-quality-gate | ECU and network | Release gate for ECU software. UDS diagnostics over ISO-TP (udsoncan, can-isotp, python-can), CAN timing and security checks run on a virtual bench, and a build with seven planted defects proves the suites catch them. |
| ecu-fault-injection | ECU on hardware | Fault injection bench for an STM32 ECU. Lost and corrupted commands, a stuck CPU, bus-off and interrupted firmware updates over CAN are injected on purpose. One test suite is written for both simulation and the board: it passes in simulation, and the run on the board is pending. |
| llm-testcase-review | Test design research | Measures LLM-written test cases by executing them on reference and defect versions, and repairs the test set with boundary, cross-check and mutation feedback. |
flowchart LR
A[automotive-sw-qa<br>AEB decision logic<br>and its test design] -- same function, on an ECU --> B[ecu-quality-gate<br>diagnostics, network,<br>security, release verdict]
A -- reference implementation<br>and human baseline --> C[llm-testcase-review<br>how good are<br>LLM-written tests?]
B -- from a virtual ECU<br>to a real MCU --> D[ecu-fault-injection<br>safety mechanisms and<br>firmware update on an STM32]
Autonomous driving
Software quality and verification
2026.01 — Present President — KUUVe · Kookmin University Unmanned Vehicle
2025.06 — Present Field Test Engineer — KUUVe Scale Car Project
2023 — Present B.E. Automotive Engineering — Kookmin University
| Date | Competition | Award |
|---|---|---|
| 2026.07 | SEA:ME Hackathon (Volkswagen Foundation) | Gold Prize |
| 2026.10 | Autonomous Robot Race, Round 3 | Excellence Award (3rd) |
| 2026.03 | 5th Int'l University EV Autonomous Driving Competition | Effort Award |
| 2025.11 | International Robot Contest — TurtleBot3 Autorace | Encouragement Award |