Probabilistic state estimation using force sensor information in contact motion of arm robot

This paper presents a method for estimating object poses based on force sensor information using a particle filter. Autonomous robots often face uncertainty of measurement of object position when they manipulate objects. Related studies dealing with uncertainty in manipulation mainly focused on motions while grasping objects. In contrast, however, a non-grasp contact motion can be advantageous because it can be achieved with simple mechanical structures. In this paper, we propose a method of estimating the object's pose in the object-contact motion without a rigid grasp. We apply a unilateral constraint to the state estimation by combining a contact force calculation model with the particle filter estimation scheme. The proposed method was evaluated using simulation, where force sensor information was effectively utilized to estimate the object's pose. Moreover, it was confirmed that the information of not sensing any force also helped the robot to narrow its state distribution when considering the interference between the robot and the object.