Grasp Synthesis for Human-Virtual Character Handovers

Abstract

We create grasps for a human-to-virtual-character handover. This is a grasp synthesis problem with the added requirement of avoiding the human hand, while aiming for natural and comfortable looking output grasps. Grasp synthesis is mainly used in the robotics domain for real life applications. Here the forces applied by the robot hand must be calculated through a precise physics model in order to correctly predict their real-world consequences. As our situation is a virtual one, we can approximate the physics requirements. Our approach is based on random sampling of hand placements in proximity of the object and then closing the fingers around the object. We analyse the parameters used in our method and constrain their ranges in order to shorten the execution time. We have made a significant improvement with respect to the naive version of the algorithm (which does not restrict the ranges of the parameters): the final version of our algorithm finds a valid grasp in 3.6% of the time taken by our naive version.