Optimal action selection to estimate the aperture of bag by force-torque sensor

Abstract

ABSTRACT In this study, we applied Particle Filter to the state estimation of deformable object based on force-torque sensor observations, and proposed information gain-based optimal action selection method. The objects that exist on the production sites, such as packed circuit boards and bags containing objects, are translucent and deformable. For these objects, it is difficult to apply the state estimation methods based on image sensor widely used in production sites. In addition, it is not possible to uniquely determine the shape of the object. In this study, Particle Filter based on observation by force-torque sensor is applied to represent the object state probabilistically. On the other hand, observation by force-torque sensor requires the robot to move actually, which takes more time than observation by image sensor, and the number of times the robot has to move should be reduced as much as possible. Therefore, in this research, we evaluate the information gain of particle distribution, and the action that can most narrow down the state of the object is the contact action with the largest information gain. The proposed method was applied to the estimation of the aperture of the packed circuit board and evaluated.