Jun-Gill Kang

Building robots that model physics, perceive the world, and move decisively in the wild.

Researcher at the Defense AI Center (ADD), specializing in physics-grounded control and real-world embodied systems.

Hero background
01

Model the machine

Physics-grounded controllers and simulators.

02

Perception for action

Vision and mapping shaped for navigation.

03

System validation

Closing the loop on real-world robots.

A broader record of systems and experiments.

01

Control & Physics simulation

Flying squirrel drone with foldable wings

01.2022~06.2025

Thrust-limited quadrotor with deployable wings, trained with residual RL.

Robot design RL Learning from Demonstration Non-linear control

KAIST HOUND

06.2022~Current

Learning based control combined with Trajector optimization.

Trajectory Optimization RL Representation learning Perception

Whole body lagrangian dynamics simulator for quadruped

06.2021~12.2021

Simulation of 12 DOF quaduped dynamics with Hierarchical QP based motion planning.

Rigid body dynamics simulation QP Advanced system control

Digital twin with 3D scanning

06.2023~06.2024

Scanning outdoor environment with iPad and develop raycasting LiDAR simulation.

Sensor simulation Digital twin Physics simulation

Kinematics & Dynamics of manipulator

02.2022~06.2022

Textbook : A Mathematical Introduction to Robotic Manipulation.

IK with manipulability maximization & Joint Limit & Null space optimization

02

Perception & Navigation

Perceptive locomotion

06.2022~Current

Full-size quadruped · outdoor terrain · vision-based control.

Perception Depth Camera 2D LiDAR Sim-to-real
Navigation foundation model visualization

Navigation Foundation model

12.2024~Current

Implement LiDAR based navigation foundation model.

VLM Large transformer training Imitation learning

LiDAR mapping

08.2023~08.2024

Flexible voxel hash mapping for LiDAR map.

Lidar-inertial odometry (LIO) ROS2 LiDAR perception
Off-road self-driving platform

Off-road self-drvining (MPPI)

08.2023~08.2024

Off-road traversal of self-driving car for military purpose.

Model predictive path integral (MPPI) Neural network dynamics LiDAR perception
Multi-robot exploration systems

Multi-robot exploration

08.2023~10.2025

Multi-robot exploration with limited communication band.

ROS2 Lidar-Inertial-Odometry Algorithms Navigation & Exploration algorithms

03

Hardware

Wheel-legged robot

Development of wheel legged robot

06.2021~12.2021

Develop wheel-legged robot with CAN protocol.

3D CAD 3D printing Simulating whole-body lagrangian dynamics Sequential QP based motion planning
Glidable drone development

Development of glidable drone

02.2022~06.2022

Develop glidable drone with foldabe wings.

3D CAD 3D printing State estimation Kalman filter
Wall-climbing robot panorama

Dual-unit wall-climbing robot

06.2020~06.2022

Ceiling adhesion robot with laser projector based inspection mission plannig.

3D Printing Physics modelling STM32 board programming

CAD Design & Arduino development

09.2020~12.2020

Developped 3D printing engine & Arduino with E-CAD

3D CAD & Rendering E-CAD

Cycloidal motor reducer development

09.2020~12.2020

Developped cycloidal motor reducer with ROS2 interface

3D CAD & Rendering ROS2

Field-ready systems reaching the public and leadership.

Media & Outreach

Public attention for bio-inspired flight

Informal interview from IEEE Spectrum, New Atlas cover, Tech Xplore cover, and major Korean press coverage (SBS, YTN) for the Flying Squirrel Drone.

Outreach demonstration

Demonstrations

Marc Raibert Demonstration

Marc Raibert (Boston Dynamics)

Demonstration of hurdle jumping controller at KAIST.

View Video
Presidential Demonstration

Presidential Visit (ADD)

System demonstration for the 14th President of South Korea.

View Video