During the last fifty years, the scientific community has dealt with the problem of designing humanoid robots, with the goal of creating artificial machines flexible enough to master any task performed by a human. A prerequisite for such generic machines would be the ability to control the forces that robots exchange with the environment, in addition to their own motion. These requirements imply that any controller of a humanoid robot requires an implicit or explicit model for the robot’s dynamics, i.e. the laws describing the relation between the robot motion and the forces applied on itself, being either the external forces that the robot exchange with the environment or the forces provided by its own motors. Additionally, any time-variant quantity present in these models need to be perceived by the humanoid robot. If a quantity is not directly measured by a sensor, it needs to be estimated using the available measurements of different quantities and the dynamical models that relate them. Both the dynamics models and the sensor models are typically not perfectly known, and need to be identified from measured data and using a set of a-priori hypothesis. This thesis focuses on addressing the problems of modelling, estimation and identification of humanoid robots, focusing in particular to the specific characteristics and sensor set of the iCub humanoid robot.