A Sensor Fusion Framework for Finding an HRI Partner in a Crowd

We present a probabilistic multi-modal system for a robot to detect and approach the most promising interaction partner in a crowd of uninstrumented people outdoors. To achieve robust operation, the system integrates three multimodal percepts of humans and regulates robot’ behaviour to approach the location with highest probability of an engaged human. A series of real-world experiments demonstrates the robustness and practicality of the proposed system for controlling robot attention.