Enabling robotic systems to collaborate with humans is a challenging task, on different levels of abstraction. Such systems need to understand the context under which they operate, by perceiving, planning and reasoning to team up with a human. The robotic system should also have perspective taking capabilities in order to efficiently collaborate with the human. In this work an integrated cognitive architecture for human robot collaboration, that aims to develop perspective taking capabilities using human preferences, is proposed. This is achieved by developing a ‘mental model’ that takes human preferences, the knowledge of the task (including the objects), and the capabilities of the human and the robot. This mental model forms the basis of the cognitive architecture, to perceive, reason and plan in the human-robot collaborative scenario. The robotic platform guided by the cognitive architecture, performs ‘picking’, ‘showing’, ‘placing’ and ‘handover’ actions on real world objects (of interest in the assembly process) in coordination with the human. The goal is to answer the ‘how’ (how a manipulation action should be carried out by the robot in a dynamically changing environment) and the ‘where’ (where the manipulation action should take place) of the assembly process considering/given varying human preferences. We show that the proposed cognitive architecture is capable of answering these questions through various experiments and evaluation.