Autonomous robots are being used in increasingly difficult real-world
situations. These exciting new applications highlight the current limitations of
autonomy. Advancing the state of the art will require significant research on
core problems in robotics, including motion estimation and motion planning.
This talk will present work on understanding these problems and using this
knowledge to design better solutions. It will present work on multimotion
estimation and motion planning in continuous search spaces. The first extends
the success of Visual Odometry (VO) to measuring other motions in the scene.
The second unifies and extends informed, graph-based search (e.g., A*) and
anytime, sampling-based planning (e.g., RRT*) to create informed, anytime
sampling-based planning algorithms for a variety of problems in continuous
spaces. These planning algorithms exploit universal properties of the problem
to perform better on many real-world robotic planning problems, especially in
the presence of complex constraints or high state dimensions.