Driven by vision. Defined by persistence.
B.S. in Automotive EngineeringKookmin University, Senior
The ultimate goal — an end-to-end autonomous driving system that unifies every layer of the stack into a single, learned self-driving pipeline. From raw sensor input to steering and throttle output, trained with reinforcement learning and imitation learning.
Building a complete end-to-end autonomous driving system that takes raw sensor data (camera, LiDAR) and directly outputs driving commands (steering, throttle, brake). Inspired by Tesla's FSD approach but built entirely from scratch with custom hardware and software.
Planning a transformer-based architecture that processes multi-modal sensor inputs through a shared backbone. The model will learn perception, prediction, and planning jointly — eliminating the hand-engineered interfaces between traditional stack layers.
A hybrid training approach: imitation learning from human driving demonstrations to bootstrap the policy, then reinforcement learning in simulation to push beyond human performance. The custom Unity simulator (Layer 06) provides the high-fidelity training environment.
Every preceding layer contributes — hardware platforms for deployment (01), perception as training signal (02), localization for ground truth (03), planning baselines for comparison (04), control as fallback safety layer (05), and simulation for scalable training (06).