Site Search  

Resources » Publication Details

Publication Details


Reference TypeJournal Article
Author(s)Stulp, F.;Theodorou, E.;Schaal, S.
Year2012
TitleReinforcement Learning with Sequences of Motion Primitives for Robust Manipulation
Journal/Conference/Book TitleIEEE Transactions on Robotics
AbstractPhysical contact events often allow a natural de- composition of manipulation tasks in terms of action phases and subgoals. Within the motion primitive paradigm, each action phase corresponds to one motion primitive, and the subgoals correspond to the goal parameters of these primitives. Current state-of-the-art reinforcement learning algorithms are capable of efficiently and robustly optimizing the parameters of motion primitives in very high-dimensional problems. Whereas these algorithms consider only the shape parameters, which determine the trajectory between the start- and end-point of the movement, in manipulation it is also crucial to optimize the goal parameters, which represent the subgoals between the motion primitives. We therefore extend the Policy Improvement through Path Integrals (PI2) algorithm to simultaneously optimize shape and goal parameters. Applying this approach to sequences of motion primitives leads to ‘sequential reinforcement learning’. We use goal learning and sequential reinforcement learning to address a fundamental challenge in manipulation: improving the robustness of everyday pick-and-place tasks
Short TitleReinforcement Learning with Sequences of Motion Primitives for Robust Manipulation
Custom 3papers2://publication/uuid/02761319-5029-4A34-880E-C0E8E6A6F34A
Link to PDFhttp://www-clmc.usc.edu/publications/S/stulp-TR2012.pdf

Designed by: Nerses Ohanyan & Jan Peters
Page last modified on June 20, 2013, at 07:00 PM