Towards Efficient and Robust Robot Planning

Abstract

In this work, three contributions are made to state-of-the-art robot planning. The contributions expand robot planning to be more efficient and robust by first expanding the mapping between task space and joint space via improved inverse kinematics. This improved mapping allows planning to be robust by increasing the size of the goal set. Second, an optimizing version of LQR-Trees is provided, this allows for high-performance and robust controllers to be constructed automatically. Finally, a new method for constructing symbolic representations with controllers that are parameterized expands the applicability of symbolic planning to a wider set of controllers.