Representation Learning for Modeling and Control of Flexible Manipulators

Traditional rigid serial manipulators have a natural representation associated with them – namely the angles of all the joints in the robot arm. A complete framework that is built around this representation allows engineers to write down simple kinematic and dynamic equations for the robot. These equations enable robust control of rigid robot arms. In contrast, for flexible manipulators such as catheters, steerable needles, and active cannula, no general modeling and control framework exists. One limitation is the lack of a natural representation of these robots. In this work, we investigate learning a representation for flexible manipulators that may be useful for modeling and control of flexible manipulators.