Adaptive Gesture Extraction and Imitation for Human-Robot Interaction

A robotic system for extracting and imitating gestures in Human-Robot Interaction is presented. Using the iCub robot and RGB-D sensors, contextually important time points are used to mark and ground the gestures. The benefits of this approach are that the gesture imitation appears both visually and behaviourally realistic, the cognitive and computation loads are reduced and no predefined mappings or speech-based annotations are needed. The system is adaptive to individual users and contributes to the creation and maintenance of a smoother interaction.