Human-robot interaction represents the cornerstone for the full development of Industry 4.0 and 5.0 paradigms, that rely on this cooperation in order to develop more efficient and flexible production lines. In this context, the human-robot handover plays a crucial role and many approaches were introduced to plan and control this task, including the less investigated decoding of human muscles activity. Hence, the design of reliable myoelectric human-robot interfaces is a point of primary interest. This paper investigates the use of a wearable device, i.e. an armband, for achieving a robust detection of several human grasping gestures. An evaluation of the most useful features, belonging to three different computational domains, is also proposed. Outcomes showed that high recognition performance can be achieved with limited computational burden, which is crucial when dealing with real-time demands in collaborative task.