Tactile-Based In-Hand Object Pose Estimation.

This paper presents a method to estimate the pose of an object inside a robotic hand by exploiting contact and joint position information. Once an initial visual estimation is provided, a Bootstrap Particle Filter is used to evaluate multiple hypothesis for the object pose. The function used to score the hypothesis considers feasibility and physical meaning of the contacts between the object and the hand. The method provides a good estimation of in-hand pose for different 3D objects.