This research addresses the challenges of recognizing human gestures when there are systematic variations in sensor outputs. It introduces two frameworks based on hidden Markov models (HMMs) designed to model and recognize gestures that vary in systematic ways. The first framework assumes that the systematic variation is communicative and that the input gesture belongs to a gesture family. It models this variation explicitly using a parametric hidden Markov model (PHMM). The second framework overcomes signal variation by relying on online learning rather than conventional offline, batch learning. These frameworks offer improved methods for gesture recognition in the presence of systematic variations. By explicitly modeling variation or employing online learning techniques, the approaches enhance the robustness and accuracy of gesture recognition systems, enabling more reliable human-computer interaction.