Machine Learning and Friends Lunch
Extending Fingertip Grasping to Whole Body Grasping
Rob Platt
UMass
Abstract
Although it is mechanically possible for a robot manipulator to grasp using non-fingertip contacts, there are few examples of this. We call non-fingertip grasping such as grasping with proximal finger phalanges or grasping an object between the palms ``whole body grasping.'' In this talk, we give two examples of whole body grasping. In the first, we show how combinations and sequences of whole body grasp controllers can generate effective whole-hand enveloping grasps. In the second, we use whole body grasp controllers to describe a two-hand manipulation sequence. For both of these cases, we are looking for ways to automatically generate grasp and motion control combinations/sequences that accomplish specific objectives. We are interested in describing classes of problems that can be addressed with the same (or similar) control policies. Finally, we are interested in identifying macro-actions that are useful for accelerating future learning. I will solicit advice from the audience about how to handle these (and other) issues.