In this PhD thesis we focus on EMG based interfaces that can be efficiently used for Human Robot Interaction (HRI) applications in structured and dynamic environments. Initially, we present a series of advanced learning schemes for EMG based interfaces that take advantage of both a classifier and a regressor, in order to split the task-space and provide better human motion estimation accuracy with task specific models.Regarding HRI applications, we mainly focus on anthropomorphism of robot artifacts. At first we distinguish between the different notions of anthropomorphism and we introduce Functional Anthropomorphism for mapping human to anthropomorphic robot motion, respecting at the same time specific human imposed functional constraints. Then we propose a methodology for quantifying anthropomorphism of robot hands, based on set theory and computational geometry methods. This latter methodology concludes to a comprehensive score of anthropomorphism that ranges between 0 (non-humanlike) and 1 (human identical) and can be used for various robot artifacts.Subsequently, we develop a series of open-source, modular, intrinsically-compliant, low-cost, light-weight, underactuated robot hands that can be easily reproduced with off-the-self materials. The proposed hands, efficiently grasp a plethora of everyday life objects, under object pose and/or shape uncertainties and can be used for various HRI applications or even as affordable myoelectric prostheses.In order to prove the efficiency of the proposed methods, we have conducted numerous experiments involving different robot artifacts, operating in both structured and dynamic environments. ; Σε αυτή την διδακτορική διατριβή επικεντρωνόμαστε σε διεπαφές ηλεκτρομυογραφικών σημάτων οι οποίες μπορούν να χρησιμοποιηθούν για εφαρμογές αλληλεπίδρασης ανθρώπου ρομποτικών συστημάτων, τόσο σε δομημένα όσο και σε δυναμικά περιβάλλοντα.Αρχικά παρουσιάζουμε μια σειρά από προηγμένα σχήματα μηχανικής μάθησης για διεπαφές ηλεκτρομυογραφικών σημάτων, τα οποία συνδυάζουν έναν ...