Title: |
Perception-Action Loops for Autonomous Navigation From Steering Needles to Driving Robots in Complex Environments |
|
Speaker: |
Dr. Vinutha Kallem |
|
Affiliation: |
University of Pennsylvania in the GRASP Laboratory, Philadelphia, PA |
|
When: |
Tuesday, November 30, 2010 at 11:00:00 AM |
|
Where: |
MRDC Building, Room 4211 |
|
Host: |
Andrea Para | |
Abstract This talk will address two questions: What if sensitive organs and anatomical obstacles prevent a physician from accessing a percutaneous target using a straight, rigid needle? What if there are obstacles blocking the path of mobile robot or a team of heterogeneous robots? While these questions arise from categorically different applications, we will demonstrate how a common framework involving perception-action loops can be used to navigate these constrained systems. One promising solution for the first question involves steering flexible tip-steerable needles. These needles introduce exciting robotics and control systems challenges because the needle tip evolves on a Lie group, and the system exhibits a high degree of nonholonomy. We present image-guided controllers for steerable needles to improve the accuracy of needle insertions. We develop nonlinear observer-based controllers to drive the needle tip to a desired subspace and use these along with subspace planners for the needle tip to reach a desired location inside the tissue. We show that the tasks of these controllers induce symmetry, thus resulting in a reduced system which greatly simplifies controller and observer design. We propose a method to perform such task-induced reduction for a broader class of nonholonomic systems on Lie groups. For the second question, mobile robot navigation among obstacles, the robots in question share similar mechanical constraints as the steerable needles. Again, we apply a decomposition approach as the basis for forming perception-action loops, although here the decomposition takes on a somewhat different flavor. Specifically, we address this question by decomposing the free workspace into cells and generating local smooth feedback laws that drive a single robot or a team of robots from one cell to an adjoining cell. These local controllers are then sequenced using discrete graph search methods like A* or incremental D* to reach the goal. We present successful implementations of the controllers in navigating ground robots and teams of ground robots and flying robots. We show that these methods can be also used for designing optimal controllers for constrained nonlinear systems. |
||
Biography Vinutha Kallem is a Research Scientist at the University of Pennsylvania in the GRASP Laboratory. Her research aims to provide autonomy in mobile robotics systems and supervised autonomy for surgical systems. Her research interests include dynamical systems, control theory, robotics, and their applications in solving real-world problems that arise in mobile systems, medicine, and health care. Dr. Kallem received her Ph.D. in Mechanical Engineering from Johns Hopkins University in 2008, where she was also a recipient of the Heath Fellowship. Dr. Kallem has an M.S. from Stanford University and a B.Tech. from the Indian Institute of Technology Madras, both in Mechanical Engineering. |