Reliable manipulation using multi-modal information: Handling of objects and door-opening

Abstract

Service robots usually handle objects, e.g., carry an object from a location to other place or work with tools in their hands. Because of recognition, estimation, and grasping error and unexpected contact with environment, there always exist errors in the pose of the object with respect to a reference frame. It is necessary to overcome such uncertainties to accomplish tasks successfully. To implement a robust and reliable manipulation algorithm that works under such conditions described above, we develop a manipulation algorithm that incorporates a pose estimator and compliance control. We adopt two pose estimators: one is based on scaling series particle filter, and the other is a vision-based marker tracking system. We also develop a new pose estimation algorithm that uses contact forces between the object and environment. Handling of an object in contact with other objects requires compliance control with respect to a proper reference point, or a compliance center. Utilizing one of the three pose estimators and a force-torque sensor, our manipulation algorithm builds forward/inverse kinematics of the robot and the object in its hand, and enables compliance control with respect to the compliance center. To show the reliability and robustness of the manipulation algorithm, we experiment a door-opening task and a puttingdown-a-cup task.