공간 구배 입력을 사용한 인공신경회로망 기반 환경 인지 (Environment Perception using Artificial Neural Network with Spatial Gradient Inputs)

서지정보

· 발행기관 : 제어·로봇·시스템학회
· 수록지 정보 : 제어.로봇.시스템학회 논문지 / 25권 / 6호 / 485 ~ 491페이지
· 저자명 : 이수용

초록

Mobile robots and autonomous vehicles are used for many areas, including exploration, delivery, rescue, and even homecleaning. The way that the robot perceives the environment plays a key role because the environmental model can be used forlocalization and mapping. Examples of environmental model generation tools are simultaneous localization and mapping (SLAM) thatwas developed for robotics and 3D maps that are usually generated from vision system data. In parallel, artificial intelligence (AI) hassupplemented human reasoning and has affected various fields as researchers developed the computer hardware and algorithms. Whilethe learning capability of AI has been used for environment perception, the success of AI is dependent on input parameter selection,structure, and training data. In this paper, we used spatial gradient information as additional inputs and introduced a perturbation/correlation-based spatial gradient estimation, which is very robust to noise since it is performed as an integration instead of the typicalapproach of calculating the gradient. Our experimental results show the effectiveness of the spatial gradient estimation. We presented anexample to illustrate the improved perception from the additional gradient inputs to the AI algorithm.

영어초록

Mobile robots and autonomous vehicles are used for many areas, including exploration, delivery, rescue, and even homecleaning. The way that the robot perceives the environment plays a key role because the environmental model can be used forlocalization and mapping. Examples of environmental model generation tools are simultaneous localization and mapping (SLAM) thatwas developed for robotics and 3D maps that are usually generated from vision system data. In parallel, artificial intelligence (AI) hassupplemented human reasoning and has affected various fields as researchers developed the computer hardware and algorithms. Whilethe learning capability of AI has been used for environment perception, the success of AI is dependent on input parameter selection,structure, and training data. In this paper, we used spatial gradient information as additional inputs and introduced a perturbation/correlation-based spatial gradient estimation, which is very robust to noise since it is performed as an integration instead of the typicalapproach of calculating the gradient. Our experimental results show the effectiveness of the spatial gradient estimation. We presented anexample to illustrate the improved perception from the additional gradient inputs to the AI algorithm.

참고자료

· 없음