ROOT ▮ ENERGY-EFFICIENT AI BRANCH_2 OF ONE SPINE
TRACK 03 — FOR HYUNDAI · TESLA FOLKS
Perception that fits
in the car.
Autonomy fails at the edge, not in the data center. My work makes perception run on automotive-grade silicon: thermal-RGB fusion that finds pedestrians cameras miss, lane detection units small enough for microcontrollers, and a GTA5 digital twin to test it all safely.
WHY ME FOR THIS TRACK
Multi-modal is my default
Camera + LiDAR + thermal. Selective thermal fusion lifted pedestrian detection from 40.4% to 83.9% — published at IEEE PerCom.
I ship to embedded targets
Jetson Nano deployments with measured FPS and memory budgets; lane detection units that fit lightweight automotive MCUs (IEEE Access).
Digital twin before the road
Built a GTA5-based digital twin interoperating with real embedded controllers to validate self-driving algorithms safely.
EVIDENCE
Thermal-RGB Fusion for Pedestrian Safety
40.43% → 83.91% accuracy · 2.7 FPS PC / 0.75 FPS Jetson Nano, 140 MB
Low-Power Lane Detection Unit for Automotive MCUs
Sliding-based parallel segment detection accelerator — journal publication
2D Image + LiDAR Fusion for 3D Object Recognition
Parallel processing pipeline for real-time 3D perception
GTA5 Digital Twin for Autonomous Driving
Game-engine simulation interoperating with embedded controllers — Electronics 2021
Perception, but
production-grade.
OPEN TO RESEARCH INTERNSHIPS — ADAS / AUTONOMY / PERCEPTION HW TEAMS