The introduction of humanoid robots into real-life environments depends on a robust and safe operation. One of the main characteristics defining a humanoid robot, as the human case, is its ability to maintain an upright bipedal stance. It is a key factor to perform manipulation tasks in real environments. This proposal aims to implement and improve different bioinspired balance strategies and controllers, while the humanoid robot performs a particular manipulation task. The behaviour proposed consists of transport an object without grasping it, similar to a human waiter transporting objects on a tray. During the execution of this behaviour, a double balancing task must be performed. The robot must keep its own equilibrium meanwhile keeping the items on the tray, that is, preserving its stability and its condition. Thanks to the capacities of Eurobench testbeds, it will be possible to evaluate robot performance in a wide variety of situations that could occur in the real world. This question constitutes an advantage in the development of balance controllers using bioinspired techniques, such as Neural Networks, Fuzzy systems, etc. The data from experiments is applied in a behaviour learning process. The richer is the information used in this process, the better the controller performance will be. The experiments will be carried out with the humanoid robot REEM-C, and the results will also be compared with the data obtained with the humanoid robot TEO from the RoboticsLab research group. The possibility of comparing Eurobench proposal results with those obtained in RoboticsLab laboratory will enrich the outcomes.