Chairs: C. Cirstea, F. Gadducci, H. Schlingloff Past Chairmen: M. Roggenbach, L. Schröder, T. Mossakowski, J. Fiadeiro, P. Mosses, H.-J. Kreowski
Fri, 01 April 2016 at 03:00 pm in Eindhoven, The Netherlands
Joint work with: Lenz Belzner and Rolf Hennicker
Abstract: In this talk we propose a modeling framework for autonomous systems operating in domains with large probabilistic state spaces and high branching factors [1]. The framework defines components for acting and deliberation, and specifies their interactions. It comprises a mathematical specification of requirements for autonomous systems. We discuss the role of such a specification in the context of simulation-based online planning. We also consider two instantiations of the framework: Monte Carlo Tree Search for discrete domains, and Cross Entropy Open Loop Planning for continuous state and action spaces. The framework's ability to provide system autonomy is illustrated empirically on a robotic rescue example. [1] Lenz Belzner, Rolf Hennicker, Martin Wirsing: OnPlan: A Framework for Simulation-based Online-Planning. In Christiano Braga, Peter Olvecky (eds.): FACS 2015, Niteroi. Springer Lecture Notes in Computer Science 9539, 2016, 1-30.
Slides