보고, 이해하고, 스스로 움직이다: 멀티모달 융합 정보 기반 자율 모빌리티
일시11/26 목 · 13:00~13:30
발표장세미나실[7]
- 저자
- 오기용, 전문수(한양대)
- 소속
- 한양대
- 좌장
- 손정우(금오공대)
연사 소개
- 2006년 한국과학기술원(KAIST) 기계공학 석사
- 2016년 미시간대학교(Univ. of Michigan Ann Arbor) 기계공학 박사
- 2007년 한국전력공사 전력연구원 선임연구원
- 2016년 ~ 2017년 미시간대학교 Research Fellow
- 2017년 ~ 2021년 중앙대학교 에너지시스템공학부 교수
- 2021년 ~ 현재 한양대학교 기계공학부 교수
강연 내용
- 멀티모달 센서 융합 기반 강건 인지 및 위치 추정 기술
- 기하·외관·열 정보 통합 다차원 공간 이해 기술
- 휠·보행 플랫폼 범용 자율 이동 및 임무 수행 기술
키워드
확장요약문
Autonomous mobility is rapidly expanding beyond structured roads into industrial sites, public infrastructure, and urban facilities shared with humans and manned equipment. However, conventional systems based on a single sensing modality degrade under unstructured terrain, multi-story structures, and adverse conditions such as darkness, smoke, and reflective surfaces. Existing multimodal approaches mostly adopt fixed fusion strategies regardless of frame-wise sensing reliability. This study proposes a multimodal fusion-based framework that enables autonomous mobility to see, understand, and move autonomously.
The proposed framework is built on three key pillars. First, robust localization is achieved by fusing 3D light detection and ranging (LiDAR), optical, thermal, and inertial measurements within a simultaneous localization and mapping (SLAM) architecture. Second, the fused information is transformed into a spatial understanding map integrating geometry, appearance, and thermal information (Fig. 1). Third, the map enables platform-agnostic navigation of wheeled and legged mobility.
Field experiments demonstrate the effectiveness of the proposed framework. Optical-assisted localization degrades at night owing to reduced visual features. In contrast, the proposed framework maintains trajectory errors of 0.140 m and 0.157 m in daytime and nighttime, respectively. In glass-dominant indoor environments, LiDAR-only SLAM yields a mapping error of 2.171 m owing to reflections. The proposed framework reduces the error to 0.377 m. Note that the perception-aware weighting suppresses unreliable measurements. These results confirm that multimodal information fusion is a key enabler of reliable autonomy across diverse operating conditions. Future work will address safe interaction with humans and cooperation among multiple mobility platforms. Ultimately, physical AI is expected to extend into everyday environments.