4족 보행로봇의 Backpropagation 신경망이론을 이용한 가상 센서 기술 검증에 관한 연구

Abstract

The sensor data measured by the quadruped robot are used to recognize the physical environment or information that controls the robot?s posture. Therefore, a robot?s ambulation can be advanced with the use of such sensing information. For the precise control of a robot, highly accurate sensor data are required, but most highly accurate sensors are expensive and are exposed to excessive load operation in the field. The seriousness of these problems will be seen if the prototype?s practicality and mass productivity, which are closely related to the unit cost of production, preservation, and maintenance, will be considered. In this paper, the use of a virtual sensor technology was suggested to address the aforementioned problems, and various ways of applying the theory to a walking robot through the use of virtual sensor information obtained through training with an actual sensor, and of various hardware information, were presented. Towards this end, the back propagation neural-network theory was used. Finally, the possibility of the replacement of the ground reaction force sensor with a leg joint encoder of a quadruped robot was verified.