Visual Servoing for Humanoid Grasping and Manipulation Tasks

Using visual feedback to control the movement of the end-effector is a common approach for robust execution of robot movements in real-world scenarios. Over the years several visual servoing algorithms have been developed and implemented for various types of robot hardware. In this paper, we present a hybrid approach which combines visual estimations with kinematically determined orientations to control the movement of a humanoid arm. The approach has been evaluated with the humanoid robot ARMAR III using the stereo system of the active head for perception as well as the torso and arms equipped with five finger hands for actuation. We show how a robust visual perception is used to control complex robots without any hand-eye calibration. Furthermore, the robustness of the system is improved by estimating the hand position in case of failed visual hand tracking due to lightning artifacts or occlusions. The proposed control scheme is based on the fusion of the sensor channels for visual perception, force measurement and motor encoder data. The combination of these different data sources results in a reactive, visually guided control that allows the robot ARMAR-III to execute grasping tasks in a real-world scenario.