Reaching and Grasping of Objects by Humanoid Robots through Visual Servoing

Visual servoing allows to control the motion of a robot using information from its visual sensors to achieve manipulation tasks. In this work we design and implement a robust visual servoing framework for reaching and grasping behaviours for a humanoid service robot with limited control capabilities. Our approach successfully exploits a 5-degrees of freedom manipulator, overcoming the control limitations of the robot while avoiding singularities and stereo vision techniques. Using a single camera, we combine a marker-less model based tracker for the target object, a pattern tracking for the end-effector to deal with the robot’s inaccurate kinematics, and alternate pose based visual servo technique with eye-in-hand and eye-to-hand configurations to achieve a fully functional grasping system. The overall method shows better results for grasping than conventional motion planing and simple inverse kinematics techniques for this robotic morphology, demonstrating a 48.8% of increment in the grasping success rate.