Efficient Continuous-Time Reinforcement Learning with Adaptive State Graphs

Gerhard Neumann, Michael Pfeiffer, Wolfgang Maass

Publikation: Beitrag in Buch/Bericht/KonferenzbandBeitrag in einem KonferenzbandBegutachtung

Abstract

We present a new reinforcement learning approach for deterministic continuous control problems in environments with unknown, arbitrary reward functions. The difficulty of finding solution trajectories for such problems can be reduced by incorporating limited prior knowledge of the approximative local system dynamics. The presented algorithm builds an adaptive state graph of sample points within the continuous state space. The nodes of the graph are generated by an efficient principled exploration scheme that directs the agent towards promising regions,
while maintaining good online performance. Global solution trajectories are formed as combinations of local controllers that connect nodes of the graph, thereby naturally allowing continuous actions and continuous time steps. We demonstrate our approach on various movement planning tasks in continuous domains.
Originalspracheenglisch
TitelMachine learning
UntertitelProceedings, ECML 2007, 18th European Conference on Machine Learning, Warsaw, Poland, September 17 - 21, 2007
ErscheinungsortBerlin, Heidelberg
Herausgeber (Verlag)Springer Verlag
Seiten250-261
ISBN (Print)978-3-540-74957-8
DOIs
PublikationsstatusVeröffentlicht - 2007
Veranstaltung18th European Conference on Machine Learning: ECML 2007 - Warschau, Polen
Dauer: 17 Sept. 200721 Sept. 2007

Publikationsreihe

NameLecture Notes in Computer Science
Band4701

Konferenz

Konferenz18th European Conference on Machine Learning
KurztitelECML 2007
Land/GebietPolen
OrtWarschau
Zeitraum17/09/0721/09/07

Fingerprint

Untersuchen Sie die Forschungsthemen von „Efficient Continuous-Time Reinforcement Learning with Adaptive State Graphs“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren