Example of a sequence in a two-person scenario. First row shows an overview of the scene, including the participants and the robot. Second row shows the robot’s field of view. The robot’s head is first initialized in a position where no face is visible (1st column), and the model uses the available detections (elbow/wrist) to find the person on the right (2nd column). The robot finds the second person by looking around while keeping the first person in its field of view (3rd column), and tracks the two people walking together (4th column).