Identifying Trajectory Classes in Dynamic Tasks - Robotics Institute Carnegie Mellon University

Identifying Trajectory Classes in Dynamic Tasks

Conference Paper, Proceedings of International Symposium on Approximate Dynamic Programming and Reinforcement Learning (ADPRL '07), April, 2007

Abstract

Using domain knowledge to decompose difficult control problems is a widely used technique in robotics. Previous work has automated the process of identifying some qualitative behaviors of a system, finding a decomposition of the system based on that behavior, and constructing a control policy based on that decomposition. We introduce a novel method for automatically finding decompositions of a task based on observing the behavior of a preexisting controller. Unlike previous work, these decompositions define reparameterizations of the state space that can permit simplified control of the system.

BibTeX

@conference{Anderson-2007-9681,
author = {Stuart Anderson and Siddhartha Srinivasa},
title = {Identifying Trajectory Classes in Dynamic Tasks},
booktitle = {Proceedings of International Symposium on Approximate Dynamic Programming and Reinforcement Learning (ADPRL '07)},
year = {2007},
month = {April},
keywords = {Trajectories, Control, Task Identification},
}