Wed, June 9
Speaker: Lillian Chang, PhD
Title: Capturing and Transferring Human Manipulation Actions
Abstract: Understanding human manipulation can provide insight into strategies for robot manipulators, character animation of hands, and clinical evaluation techniques. In this talk I will present two research projects from my work on analysis and synthesis of manipulation actions. First, to capture accurate human hand shape for example-based methods, we need reliable methods for automatic estimation of anatomic joint parameters. In one application, we have contributed an improved method for determining the joint center of rotation using constrained least-squares optimization. The carefully selected singularity of the mathematical formulation results in an accurate and repeatable solution that is robust to the challenges present in real motion capture of hand joints. Second, human manipulation provides compelling examples of how robots could interact with the environment. We have found that humans use a "pre-grasp interaction" manipulation strategy for grasping heavy or hard-to-reach objects. We have developed a technique for robot pre-grasp interaction and have demonstrated that this strategy improves the robot's effective functionality for reaching and transporting household objects.
Host: Sandro Mussa-Ivaldi