Bidirectional Learning of Handwriting Skill in Human-Robot Interaction

This paper describes the design of a robot agent and associated learning algorithms to help children in handwriting acquisition. The main issue lies in how to program a robot to obtain human-like handwriting and then exploit it to teach children. We propose to address this by integrating learning from demonstrations paradigm, which allows the robot to extract a task index from intuitive expert (e.g., adults) demonstrations. We present our work on the development of an algorithm, as well as its validation by learning compliant robotic writing motion from the extracted index. Also discussed is the synthesis of the learned task in the prospective work of transferring the task skill to users, especially in terms of learning by teaching. The undergoing work about the design of a sensor-embedded pen is introduced. This will be used as an intuitive interface for recording various handwriting related information in the interaction.