Despite the social motivation for the use of a teachable agent for the engagement and support of children with handwriting difficulties, to date no such technology is known to have been developed. Presented is a teachable robotic agent in this context of handwriting skills acquisition. It is hypothesised that such a system could not just engage an unmotivated student, but could also present the opportunity for children to experience motor mimicry during handwriting intervention. It also allows for exploring the potential for the learning by teaching paradigm to be employed in the interaction so as to stimulate meta-cognition, empathy and increased self-esteem in the child user.
By leveraging simulated handwriting on a synchronised tablet display, a Nao humanoid robot with limited fine motor capabilities has been configured as a suitably embodied handwriting partner. Shape models derived from principal component analysis of a dataset of letter trajectories are generated and allow the robot to draw purposefully deformed letters. As a result, the objective of developing an algorithm learning how to write characters well may be satisfied by learning the optimal parameters for the model of said characters. Such a learning algorithm has been developed, capable of incorporating feedback from users both in terms of which generated letters are the best and from user demonstrations.
A pilot study, conducted to obtain insight into children’s use of the system, is presented, which has validated the interaction for an in situ experiment scheduled with a primary school class to evaluate the human-robot interaction outcomes of the system, the first of its kind.