I am a Ph.D. candidate at the Department of Aerospace Engineering, Seoul National University, Republic of Korea. My research interest includes navigation systems for autonomous robots with onboard sensors.

Most of my work is based on Bayesian filtering that fuses measurements from an inertial measurement unit, cameras, and LiDAR. My goal is to develop a practical and robust navigation algorithm from a thorough theoretical background. CV

We introduce a Gaussian mixture merge method in a matrix Lie group called "Midway-merge" that reduces an approximation error when merging probability density functions. We demonstrate the midway-merge method in an object-level SLAM problem with pose ambiguity to deal with multiple hypotheses. 

We open-source our work on GitHub. 

We formulate the invariant extended Kalman filter in the object SLAM problem. The idea is to model the state-space in the matrix Lie group and fuse visual and inertial measurements in a fully probabilistic fashion to increase the temporal consistency of a single view pose detector. We adopt a deep neural network-based pose detector as a sensor and validate our proposed method in the KITTI dataset.

We designed a pose tracking filter using RGB images, a stream of events from a dynamic vision sensor, and IMU. Our contribution is seamless fusion of multi-modal sensors in high-dynamic scenarios that is one of the most challenging problem in vision-based navigation.

Collaborative work with Hyundai MnSOFT. We developed visual-inertial-wheel odometry that replaces the noisy integration of an accelerometer with the single integration of a wheel odometer. We evaluated our system in tens of kilometers driving dataset collected in tunnels and streets in Seoul.