When a robot collaborates with people in groups, its interaction with humans is expected to be fluent and efficient. Augmenting a robot with the capacity to understand the activities of the people it is collaborating with (with specific reference to the timing of those activities), allows the robot to leverage its understanding to generate an efficient and collaborative plan to perform its actions. In this paper, we present a supervised activity segmentation algorithm that can detect the start and end time of activities, simply by observing a portion of the initial trajectory data: an essential first step in generating an efficient interaction plan for a robot. We validated the algorithm by applying it to a collaborative task involving a single robot and a single human. Results of this study indicate that the algorithm accurately segments the activities in real time with approximately 80% accuracy (partially observing the full trajectory of a given activity).
