Abstract.  Intelligent motion control lies at the core of robotic autonomy. An essential requirement for truly autonomous vehicles is their ability to efficiently navigate through natural environments. Yet, the basic motion control problem often depends on high-dimensional nonlinear dynamics, complex constraints and optimization criteria, and in general becomes computationally intractable and numerically sensitive.

In view of these challenges, we develop a computational framework that identifies and exploits structure in the problem for improving the numerical stability and efficiency of motion control. This is accomplished by: 1) discrete geometric mechanics and optimal control methods that respect and exploit inherent dynamical system properties;

2) randomized motion planning that incrementally exploits collected information through adaptive sampling. The developed methods lead to practical algorithms to advance the area of robotic mobility. As a result, a multi-layer control framework is implemented that supports dynamics simulation, feedback control, trajectory optimization, and motion planning of general constrained mechanical systems. The framework also enables high-level tasks such as active sensing and information gathering. The goal is to provide a principled approach for integrating actuators, sensor, and novel mechanism designs to realize more agile and efficient autonomous systems. We will illustrate the methods with two applications under development: reconfigurable space structures composed of small robotic spacecraft; unmanned arial vehicles navigating optimally through obstacle terrains.

Bio. Marin Kobilarov is a post-doctoral fellow in Control and Dynamical Systems at the California Institute of Technology and is affiliated with the Keck Institute for Space Studies. His current research focuses on computational control methods that exploit the geometric structure of nonlinear dynamics, on approximation methods for optimization and motion planning, and on designing and building robotic embedded systems to realize new concepts. He obtained a Ph.D. from the University of Southern California in Computer Science (2008) and a B.S. in Computer Science and Applied Mathematics from Trinity College, Hartford, CT (2003).