Kagan Tumer's Publications

Display Publications by [Year] [Type] [Topic]


Neuroevolutionary Control of a Multi-linked Inverted Pendulum with a Large Number of Agents. S. Sills and K. Tumer. In AAMAS-2012 Workshop on Autonomous Robots and Multirobot Systems, Valencia, Spain, June 2012.

Abstract

The multi-link inverted pendulum is a well known benchmark problem in control systems due to its variability in complexity. Such a system is useful for testing new learning algorithms or laying the foundation for autonomous control of more complex devices such as robotic spines and multi-segmented arms which currently use traditional control methods or are operated by humans. It is often easy to view these systems as single agent learners due to the high level of interaction among the segments. However, as the number of links in the system increases, the system becomes harder to control. This work proposes replacing the centralized learner with a team of coevolved neural networks. Using transfer learning, a team of networks can be trained to control a system with a significant number of links - 10 or more. The results presented show that using distributed control in this manner allows for control of a larger number of links than a centralized network and makes the system more robust to interference.

Download

(unavailable)

BibTeX Entry

@incollection{tumer-sills_mabs12,
        author = {S. Sills   and K. Tumer},
        title = {Neuroevolutionary Control of a Multi-linked Inverted Pendulum with a Large Number of Agents},
        booktitle = {AAMAS-2012 Workshop on Autonomous Robots and Multirobot Systems},
	month = {June},
	address = {Valencia, Spain},
	editors = {G. Kaminka and K. Hindriks},
	abstract={The multi-link inverted pendulum is a well known benchmark problem in control systems due to its variability in complexity. Such a system is useful for testing new learning algorithms or laying the foundation for autonomous control of more complex devices such as robotic spines and multi-segmented arms which currently use traditional control methods or are operated by humans. It is often easy to view these systems as single agent learners due to the high level of interaction among the segments. However, as the number of links in the system increases, the system becomes harder to control. This work proposes replacing the centralized learner with a team of coevolved neural networks. Using transfer learning, a team of networks can be trained to control a system with a significant number of links - 10 or more.  The results presented show that using distributed control in this manner allows for control of a larger number of links than a centralized network and makes the system more robust to interference.},
	bib2html_pubtype = {Workshop/Symposium Papers},
	bib2html_rescat = {Multiagent Systems, Robotics},
        year = {2012}
}

Generated by bib2html.pl (written by Patrick Riley ) on Tue Jun 26, 2018 19:10:42