Foundations of Robotics Seminar, April 21, 2011
Time
and Place | Seminar Abstract
Goal. Set. CHOMP!
Anca Dragan
Robotics Institute
Carnegie Mellon University
Thursday, April 21, 2011
NSH 3305
Talk 12:00 pm
My work focuses on robots performing manipulation tasks in
unstructured human environments, with or around humans. Such robots
need to move naturally and predictably, calling for motion planners
that optimize such criteria. While trajectory optimizers like CHOMP
(Covariant Hamiltonian Optimization for Motion Planning) have proven
effective for some high-dimensional motion planning problems, finding
feasible and optimal solutions remains hard in a lot of cases. In this
talk, I will present a way of improving trajectory optimization by
taking advantage of affordances in the manipulation problem, such as
the existence of entire regions of allowable goals. If time allows, I
will illustrate how these affordances can be used not only to improve
the optimization process itself, but also to initialize the optimizer
with better starting points.
Anca Dragan is a second year PhD candidate in the Robotics Institute,
advised by Siddhartha Srinivasa and Chris Atkeson. She was born in
Romania and received her Bachelor's degree in Computer Science from
Jacobs University Bremen, Germany. She is interested in learning from
experience and demonstration, and how they apply to manipulation in
human environments.
The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.