Object pose estimation by iterative contacts with soft tactile sensor

In order for a robot to manipulate an object, it is required to estimate the object's pose. The Manifold Particle Filter, an extension of the widely used Particle Filter as a state estimation method, was proposed for the object pose estimation. The Manifold Particle Filter can utilize tactile information, but it has only been applied to binary observations. In this paper, Continuous-MPF, which can use multidimensional and continuous observation information, is proposed. Additionally, we propose an extension of Continuous-MPF, which can store and utilize non-contact information in order to improve the efficiency of estimation. The proposed method was implemented in a robot system and experimentally validated. Good estimation results were obtained for most of the given conditions.