Playing ‘Sticky Hands’ With A Humanoid Robot

‘Sticky Hands’ is an exercise for two participants requiring them to maintain a gentle hand to hand contact. The intention is to explore graceful and relaxed motion together, develop physical sensitivity and responsiveness, and interact at a comfortably intimate level for physical relaxation and spiritual development. We present a system allowing a person to perform this novel interaction with a humanoid robot. A learning algorithm is presented which observes and predicts the trajectory of the contact point, using a novel internal representation of instantaneously sampled data ‘prototypes’. It is capable of generalizing observed trajectories to unseen cases, operating with parameterized time and memory bounds, and coping with the evolving nature of contact trajectories. We thus present a new role for the robot as a playmate for self-development, and hope to facilitate comfortable physical interaction and encourage its perception as a human-like and emotional being. We discuss augmentation of the robot's motion with human-like style and emotional expression, based on perceptual psychology, human motor production, and computer graphics -fields whose relevance for robotics is also discussed.