Logo Logo
Help
Contact
Switch language to German
Simulation-based autonomous systems in discrete and continuous domains
Simulation-based autonomous systems in discrete and continuous domains
Highly dynamic, probabilistic and potentially only partially known domains render classical techniques for system development infeasible. Enabling autonomous adaptation by providing decision making and learning capabilities yields systems with the abilities that are necessary to deal with these challenges. The key to system autonomy is to drop exact specification of runtime behavior in favor of a whole space of solutions. This space can be explored by the system at runtime according to its current situation, and potential traces in this space can be evaluated w.r.t. the system's goals. Concrete behavior is then compiled based on the results of search and evaluation. This thesis studies the use of simulation-based Monte Carlo methods for decision making and learning that enable efficient and adequate system autonomy in discrete and continuous domains.
Not available
Belzner, Lenz
2016
English
Universitätsbibliothek der Ludwig-Maximilians-Universität München
Belzner, Lenz (2016): Simulation-based autonomous systems in discrete and continuous domains. Dissertation, LMU München: Faculty of Mathematics, Computer Science and Statistics
[thumbnail of Belzner_Lenz.pdf]
Preview
PDF
Belzner_Lenz.pdf

3MB

Abstract

Highly dynamic, probabilistic and potentially only partially known domains render classical techniques for system development infeasible. Enabling autonomous adaptation by providing decision making and learning capabilities yields systems with the abilities that are necessary to deal with these challenges. The key to system autonomy is to drop exact specification of runtime behavior in favor of a whole space of solutions. This space can be explored by the system at runtime according to its current situation, and potential traces in this space can be evaluated w.r.t. the system's goals. Concrete behavior is then compiled based on the results of search and evaluation. This thesis studies the use of simulation-based Monte Carlo methods for decision making and learning that enable efficient and adequate system autonomy in discrete and continuous domains.