ETRI-Knowledge Sharing Plaform

ENGLISH

성과물

논문 검색
구분 SCI
연도 ~ 키워드

상세정보

학술대회 Anomaly Detection for Robotic Assembly
Cited 0 time in scopus Download 12 time Share share facebook twitter linkedin kakaostory
저자
김혜진, 박인준, 한효녕, 손지연
발행일
202210
출처
International Conference on Information and Communication Technology Convergence (ICTC) 2022, pp.1667-1670
DOI
https://dx.doi.org/10.1109/ICTC55196.2022.9952585
초록
Anomaly detection has developed by leaps and bounds thanks to the rapid development of computer vision in many industrial and robotic applications. In this paper, we address an anomaly detection method for assembly tasks using a robot manipulator based on deep reinforcement learning. The anomaly detection method provides reward values for reinforcement learning to estimate the assembly state: the task's success or failure. Our assembly tasks are three: Snap-fit, Peg-in-hole, and Snap-fit. Our approach is a task-agnostic method. We achieved 99.13% accuracy with 0.9799 precision, 0.9946 recall and 0.9986 AUROC based on our dataset.
KSP 제안 키워드
Computer Vision(CV), Deep reinforcement learning, Detection Method, Leaps and bounds, Peg-in-hole, Rapid development, Reinforcement Learning(RL), Robot manipulator, Robotic assembly, anomaly detection, robotic applications