Human locomotion shows fascinating abilities which are the results of the interplay between the environment, the biomechanics, the spinal cord, and modulation from higher control centers. How the different structures interact to generate meaningful behavior is an active field of research, and understanding the key principles underlying bipedal locomotion could have a strong impact and important implications in several fields related to both medicine and robotics, such as improved rehabilitation procedures, predicting surgery outcome or facilitated human-robot interaction. In this context, the development of biologically relevant bipedal models that faithfully recapitulate human locomotion are urgently needed. Existing such bio-inspired models usually rely on one of the two following principles: the Central Pattern Generators (CPGs) and the reflexes. In the first part of the thesis, we present a method to introduce a CPGs as feedforward components in a feedback based (i.e. reflex) model of human walking, named neuromuscular model (NMM). The proposed strategy is based on the idea that, in a feedback driven system, the feedforward component can be viewed as a feedback predictor. We implement the feedback predictors using morph oscillators as abstract models of biological CPGs. Thanks to the intrinsic robustness inherited from the feedback pathways, modulation of CPGs network's frequency and amplitudes becomes possible, over a broad range, without affecting the overall walking stability. Furthermore, the modulation of the CPGs network's parameters allowed smooth and stable gait modulation (such as changes in speed and adaptation to increasing slope) suggesting that the idea of using feedback predictor as gait modulator can be extended to a large range of applications, highlighting the role biological CPGs could play on top of a reflex-based rhythmic movement. Building on the NMM, we present, in the second part of the thesis, the implementations of the models as controllers on different ortheses and exoskeletons. Wearable devices designed to assist abnormal gaits require controllers that interact with the user in an intuitive and unobtrusive manner. Here, we rationalized that such a neuromuscular controller could be implemented based on the NMM models. The implementation of NMM model on a controller (NMC) was demonstrated for human healthy subject and was confirmed with experiment on SCI subjects with different devices. Overall, the bio-inspired NMCs successfully demonstrated remarkable versatility in generating gait patterns tuned to the subjects' dynamics and producing near-physiological gait at near-normative speeds. The positive SCI subject-machine interaction stemmed from replacing the subject's impaired function with dynamical virtual muscles that require few sensors. These preliminary but auspicious results have important implications towards the exploitation of natural walking dynamics through understanding human biological behavior in the design of controllers for wearable devices that are amenable to various environmental conditions and promote intuitive and unobtrusive human-machine interaction.