Understanding Agreement in Giver and Receiver Intentions on Grasp in Human-Human Handover

Fluency of handover is an essential component of human-human and human-robot collaboration. A number of approaches use multi-person handover studies to draw conclusions that inform human-robot handover behavior. Most human-human handover studies consist of objective measurements of motion or biofeedback, and lack subjective giver and receiver perceptions. Using data from a large-scale study on handover using 204 objects distributed amongst 48 human-human dyads created from amongst 32 recruited subjects, we observe that in 27.64% of grasps, the intentions of giver and/or receiver on grasp are not met, indicating lack of success in the handover for over a quarter of the cases. Our findings suggest that robotic algorithms that learn human-human handover data should be trained to demonstrate cognizance of actual versus intended behavior to ensure fluency of handover.