In recent years, the field of robotics has played a major role in the quest for restoring mobility to patients who suffered a limb impairment. Particularly, the literature in the field of bipedal robotics has provided scientific support for understanding the biomechanical interaction between artificial lower-limbs and disabled people. However, after several decades of technological progress in actuators and sensors of biped robots, there is still not enough understanding how to mimic the dexterity and efficiency of human bipedal locomotion. There is a clear open problem in the feedback control field that poses important challenges for decoding and reconstructing the fundamental biological behaviour embedded in nature. In this context, the thesis introduces a novel control architecture that is based on contraction theory and synchronisation. Combining decentralized multiple nonlinear controllers (synchronised by a virtual dynamic system) creates a mathematical abstraction of the human locomotion control. This thesis is meant to prove stability of a robust controller for an anthropomorphic walking robot. This approach is designed to minimise the risk of falling during hard joint constraints. The outcome of this research is intended to support control technology of walking assistants (e.g., exoskeletons) for patients with abnormal motor pathology (e.g., spasticity disorder due to post-stroke condition). Both theoretical and validation work presented in this thesis outperformed the results expected.