Don’t Over Think It: Mechanically Intelligent Manipulation

Abstract

Developing capable, robust robots that can effectively operate in unstructured environments requires reasoning over how embodiment and cognition impact the capability and complexity. In this thesis, we focus on the design of hands, finding a balance between capability and complexity through the lens of mechanical intelligence: the ability of the body to contribute to functionality. We develop two grippers that serve as examples of this idea, Belt Orienting Phalanges and Flexible Robust Observant Gripper.

Belt Orienting Phalanges (BOP) enables in-hand manipulation through the addition of two belts on each finger of a parallel-jaw gripper, allowing control over the roll, pitch and a translation of a grasped object. We demonstrate how these motion primitives and other aspects of BOP’s morphology enable the simple adaption of an existing planning frameworks to perform a complex manipulation tasks.

Flexible Robust Observant Gripper (FROG) eases perception and control through the structure of each finger, allowing for proprioception and robust grasping while being strong and remaining comparable in complexity to other soft grippers. We demonstrate how these features enable FROG to grasp gently and deal with shape and pose uncertainty.