Research #
3D Scene Understanding #
2021 – Present
Generative scene prior for 3D perception
Deep generative models, such as GANs and diffusion models, have demonstrated their potential as a powerful prior for data restoration tasks. Can we leverage these achievements in robotic perception? To this end, this work explores building deep generative models for LiDAR point clouds, crucial sensory data for autonomous mobile robots.
[ICRA'24] [WACV'23] [IROS'21]
Deep generative models, such as GANs and diffusion models, have demonstrated their potential as a powerful prior for data restoration tasks. Can we leverage these achievements in robotic perception? To this end, this work explores building deep generative models for LiDAR point clouds, crucial sensory data for autonomous mobile robots.
[ICRA'24] [WACV'23] [IROS'21]
2022 – Present
Sim2Real LiDAR transfer
Building a large-scale LiDAR dataset for 3D scene understanding is both expensive and time-consuming due to the need for manual annotation. While recent LiDAR simulators provide high-quality labeled pseudo-data for free, synthetic data significantly differs from real-world data. We aim to bridge this gap by adapting synthetic data to real-world data using domain adaptation techniques.
[WACV'23]
Building a large-scale LiDAR dataset for 3D scene understanding is both expensive and time-consuming due to the need for manual annotation. While recent LiDAR simulators provide high-quality labeled pseudo-data for free, synthetic data significantly differs from real-world data. We aim to bridge this gap by adapting synthetic data to real-world data using domain adaptation techniques.
[WACV'23]
2015 – 2020
LiDAR-based scene classification
[IJRR'19] [Advanced Robotics'18] [SII'17]
[IJRR'19] [Advanced Robotics'18] [SII'17]
Human–Robot Symbiosis #
2014 – 2020
Forth-person sensing in human–robot symbiotic space
[ROBOMECH Journal'20] [SMC'18] [IEEE Sensors'15]
[ROBOMECH Journal'20] [SMC'18] [IEEE Sensors'15]
Miscellaneous #
2021 – Present
Spectrum-based soil stiffness evaluation
[ISARC'23]
[ISARC'23]
2017 – 2020
Multimodal terrain classification
Recognizing terrain types plays a important role for outdoor mobile robots such as planetary rovers to estimate the traversability. We have developed a method of robust semantic segmentation from the visible and thermal imagery. This is a part of my intern work at NASA/Caltech JPL from 2017.
[RA-L'20] [CVPRW'19] [MIPR'19] [WAC'18]
Recognizing terrain types plays a important role for outdoor mobile robots such as planetary rovers to estimate the traversability. We have developed a method of robust semantic segmentation from the visible and thermal imagery. This is a part of my intern work at NASA/Caltech JPL from 2017.
[RA-L'20] [CVPRW'19] [MIPR'19] [WAC'18]
2015 – Present
Gait recognition
Gait recognition is a kind of biometrics techniques that identifies individuals based on their walking patterns. Our group explores the methods robust to illuminance and appearance changes.
[Access'24] [Access'23] [SII'22] [SII'22]
Gait recognition is a kind of biometrics techniques that identifies individuals based on their walking patterns. Our group explores the methods robust to illuminance and appearance changes.
[Access'24] [Access'23] [SII'22] [SII'22]