Humanoid robot research is one of the most challenging topics in robotics. The research in this field has been inspired by the fact that these machines replicate the human shape and, in the end, human behaviour. Many humanoid robots have been presented in the last decades, but their behaviour is far from human performance. Therefore, the improvement of humanoid performance is one of the main goals in the field, mimicking as much as possible human tasks.

This Thesis has been focused on the development of one of these behaviours and the study of open questions related to its performance. Concretely, the main goal of this thesis is to integrate a waiter behaviour in the humanoid robot TEO from the Carlos III University of Madrid. This behaviour consists of transporting objects on a tray, imitating the human task of a waiter, but this research must face different problems.

The first problem to deal with is the balance of the robot. One of the essential characteristics of a humanoid robot is its capacity of upright stance and walking. Humanoid robot balance has been one of the main problems studied from the beginning of the development of this kind of robots. Keeping balance is still a matter of study in humanoid robotics, and it is crucial to develop tasks. Due to the complexity of this problem, its study is favoured using simple representations of the humanoid balance behaviour, such as the simple inverted pendulum. This kind of simplifications makes the study of the balance problem computationally easier, but it does not solve all questions related to balance. This Thesis has dealt with balance problem because it should be a key feature for a waiter robot. A new simplified model has been proposed as an improvement of the classic Linear Inverted Pendulum Model (LIPM). Its development is based on an experimental procedure that allows tackling many inherent problems of humanoid mechanics, electronics, computing, and more at the same time. The results show that this improvement reduces the control effort need for keeping balance during the waiter task.

The second problem faced in this Thesis is related to manipulation that is another essential characteristic defining the human being: the capacity of manipulating objetcs. In this Thesis, a specific case of manipulation has been treated. In the waiter task performance, usually the objects are transported on a tray, and no gripping force is applied to them. Then another problem arises: to transport the objects keeping their equilibrium or balance. Thanks to the use of exoceptive perceptions of the robot (vision, force, etc.) are possible to interpret the balance status of the objects transported. This sensorial information and its application with a simple inverted pendulum model enable the balance control of the object.

Finally, the results obtained from the study of the two main problems exposed have been integrated into a Whole-Body Postural Control architecture. It accomplishes with all the requierments needed for performing the waiter task.