This document presents an algorithm to recognise and locate known bulky objects from a workspace and compute grasps using two anthropomorphic hands. Once valid grasp points have been found and they are kinematically reachable by the manipulator, motion planning and a collision check are performed. In the first place, this document presents a general architecture of the system and the hardware devices and software implemented. Then, a detailed description of the object recognition, grasp and motion planning and manager modules are provided. Robot Operating System (ROS) is the framework chosen to be responsible for the communication between nodes with appropriate messages, to manage the data from the sensors and to execute robot tasks. In the second place, attention is focused on the evaluation of the object recognition and grasp system. From this chapter, the most important conclusions are: - The camera’s calibration takes an important role in order to obtain good performance and reliability of the system. - This project proposes an approach to use multiple camera views to construct a final point-cloud scene to obtain a complete 3D object. - This project presents a satisfactory approach of filtering the desired object with its colours instead of using its shape. - The more complete the point-cloud is, the more reduced is the error of object pose. - A system which grasps a bulky object with six points, three for each hand, instead of four as other works recommend, is a more robust system. - The system has a bimanual manipulator with 22 DoF for each arm. Therefore, inverse kinematic problem is a laborious process to converge to a feasible solution. - ROS service is less robust to service provider changes or server failures respect to the ROS action which provides feedback on the task progress and cancelation at any time.