Jaehwan Jeong

Jaehwan Jeong

Ph.D. Student

CV Lab, Korea University

ex-VGR @ SCI Lab, UCLA

I am a Ph.D. student at Korea University, advised by Prof. Sangpil Kim. My research explores the internal structures of AI models to solve challenges across diverse domains, spanning autonomous systems for vehicles and robotics, multimodal generative models, and AI safety.

Previously, I served as a Visiting Graduate Researcher at University of California, Los Angeles (UCLA), advised by Prof. M. Khalid Jawed. In this role, I led a Smart Farm project team, driving the end-to-end deployment of robotics—specifically UGVs and manipulators—from mechanical machining and sensor integration to software, AI, and field operations.

Embodied AI Robotic Learning Autonomous Systems Generative Model AI Safety

Contact

  • jhwan@korea.ac.kr
  • Woojung Hall of Informatics, Korea University,
    Seoul, Republic of Korea

Positions

Recent News 🎉

Projects

HiVLA
VLA · RL · Robotics

Collision-free Path Recovery via VLA/RL

Detecting path deviation via internal attention heads of a frozen VLA model. Local obstacle avoidance RL enables stable navigation and direct path recovery. Full pipeline implemented with ROS2 and LIVO for real-world validation.

Robotic Pollination
Robotics · Simulation

Vision-guided Robotic Pollination

An eye-in-hand RGB-D sensor enables 3D modeling and ML-based skeletonization of plant stems. Physics simulation then determines the optimal grasp point and vibration amplitude for end-to-end autonomous robotic pollination.

AgriChrono
Robotics · Dataset · 3D Benchmark

Agricultural Dataset/Benchmark

A large-scale multi-modal dataset using a custom field robot to capture continuous crop growth and severe lighting variability. Provides a rigorous in-the-wild benchmark to drive robust 3D reconstruction models for agriculture.

FaceShield
AI Safety · Deepfake

Protect Face from Deepfake Threats

A proactive defense framework that protects facial images from deepfake threats by embedding imperceptible adversarial perturbations, disrupting synthesis without visible degradation.

A2V
Audio-to-Video · Multi-Object Generation

Multiple Object Video Generation

Audio-to-video generation for semantically complex scenes using audio source separation, mapping individual sound sources to corresponding visual content.

MEVG
Text-to-Video · Long Video Generation

Multiple Events Video Generation

Generates videos depicting multiple sequential events from text descriptions by leveraging text-to-video diffusion models with event-aware temporal control.

Publications

Academic Services

Conference Reviewers

Journal Reviewers