Locally Optimal Planning in High Dimensions - Robotics Institute Carnegie Mellon University

Locally Optimal Planning in High Dimensions

Master's Thesis, Tech. Report, CMU-RI-TR-07-34, Robotics Institute, Carnegie Mellon University, August, 2007

Abstract

Algorithms which guarantee to find globally optimal trajectories by covering a hypervolume of state space become infeasible in high dimensions. We have developed an algorithm which generates locally optimal trajectories without covering a hypervolume which only requires polynomial memory and time. Even though the algorithm does not guarantee a globally optimal trajectory in a deterministic manner, it is experimentally shown that resulting trajectories are reasonable for pendulum swing-up tasks. Results are shown for pendulum with 1- and 2-links and compared to globally optimal trajectories obtained by dynamic programming. Furthermore, results for 3-link pendulum of which globally optimal trajectory is currently unknown are presented.

BibTeX

@mastersthesis{Usui-2007-9795,
author = {Kei Usui},
title = {Locally Optimal Planning in High Dimensions},
year = {2007},
month = {August},
school = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-07-34},
keywords = {Optimal control, motion},
}