Viewpoint Selection Strategy for a Life Support Robot

Abstract

Technological advances in robotics are being made to assist people who are aged to live independently. A life support robot must be provided with the ability to pick and place intended daily objects so that tasks such as serving drinks and food could be performed successfully. Therefore, reliable six-dimensional object pose estimation is necessary. In spite of the significant research progress, the accurate estimation under partial occlusions remains a great challenge due to the limited camera position. In this work, we propose a viewpoint selection strategy for a life support robot so that it is able to actively adjust the camera's position for better viewpoints with less partial occlusions. Eventually, a viewpoint with acceptable estimation accuracy could be obtained. The presented approach has been evaluated on our life support robot KUT-LSR.