Jaehwi Jang

I am an MS student in the Graduate School of AI at the Korea Advanced Institute of Science and Technology (KAIST), advised by Prof. Daehyung Park.

My research interest lies in robust robotic skill learning, with a focus on creating robots that are user-friendly and require minimal instruction for everyday individuals. My latest projects involves learning constraints from the demonstrations for safe skill learning.

I received my bachelor's degree from KAIST in 2021, where I majored in Mechanical Engineering and Physics, and minored in Computer Science. I had the honor of being awarded the KAIST Presidential Fellowship.

Description

Honors and Fellowships

🗽Fulbright U.S.-Korea Presidential STEM Initiative Award, 2024~

Best Undergraduate Student Award, 2022
KAIST Presidential Fellowship, 2016~

Publications
Inverse Constraint Learning and Generalization by Transferable Reward Decomposition
Jaehwi Jang, Minjae Song, Daehyung Park
IEEE Robotics and Automation Letters (RA-L), 2024
[PDF] [Video] [Site (ongoing)]
SGGNet2: Speech-Scene Graph Grounding Network for Speech-guided Navigation
Dohyun Kim*, Yeseung Kim*, Jaehwi Jang*, Minjae Song*, Woojin Choi, Daehyung Park
(*- authors contributed equally)
IEEE Int'l Conf. on Robot and Human Interactive Communication (RO-MAN), 2023
[PDF]

Projects

I participated in research projects on interactive learning for robust skill intelligence and previously in a course project developing Robotics/ML projects.
Sim2Real / Real2Real for robotic skill learning
Project @ RIRO Lab, 2023

In this project, we train robotic agents to acquire human-like skills through interactive learning. We utilize simulated environments in IsaacSim or employ teleoperation to guide and teach the robots, such as deformable object manipulation.
Physics-Informed Gradient Regularization for Inverse Reinforcement Learning
RSS Workshop on Experiment-oriented Locomotion and Manipulation Research, 2023

In this project, I proposed a novel approach by integrating Physics-Informed Neural Networks (PINN) into an Inverse Reinforcement Learning (IRL) framework. This methodology utilized PINN to solve the Hamilton-Jacobi-Bellman differential equation, achieving two main objectives: (i) replication of the demonstrated behavior and (ii) recovery of the reward.
Autonomous Ball-picking Vehicle
Capstone Design, 2018

Undergraduate Capstone project at KAIST. We built an autonomous ball-collecting vehicle designed to pick up blue balls while avoiding red balls.
In this project, I participated as a group leader and was responsible for programming and control tasks.
Human Powered Aircraft
Human Powered Aircraft Competition, 2013

Participated in the 2013 HPAC hosted by Korea Aerospace Research Institute (KARI). We built a human-powered aircraft that successfully flew approximately 20 meters.
In this project, my responsibilities were specifically focused on designing and making the wings.

Additional Information

Email: wognl0402@kaist.ac.kr

List of things I have worked with:

Robots: Franka Emika Panda, UR5e, ShadowHand Lite, and RBQ-3 from Rainbow Robotics
Languages: Python (pytorch, JAX), C, C++, MATLAB and Rust
Simulators: Box2d, PyBullet, and IsaacSim

The homepage template is a modified version of Jon Barron's website
Letter j icons created by Hight Quality Icons - Flaticon