TITLE: Decision - Theoretic Planning with non-Markovian Rewards -- A Model-Checking Approach

PRESENTER: Dr Sylvie Thiebaux

AFFILIATION: Research School of Information Sciences and Engineering The Australian National University

DATE: Tuesday 8 October 2002

TIME: 2:00pm to 3.00pm

PLACE: CSE K17 1st Floor Seminar Room

ABSTRACT:

Artificial intelligence planning deals with the problem of choosing the actions to be performed to make the state of a discrete system evolve so as to satisfy given objectives. AI planning is related to e.g. supervisory control of discrete-event systems and reinforcement learning. Decision-theoretic planning is a particular case, where actions are stochastic, objectives are encoded by associating real-valued rewards to the desired state trajectories, and we seek to act so as to maximise reward. Decision processes with non-Markovian rewards (NMRDPs) form an adequate model for decision-theoretic planning. In this talk, we will examine the problem of representing and solving these NMRDPs and will present an approach making use of linear temporal logic and model-checking. No familiarity with MDPs and temporal logic will be assumed. A paper reporting work on this topic can be found at http://csl.anu.edu.au/~thiebaux/papers/uai02.pdf

BIOGRAPHY OF SPEAKER:

Sylvie Thiebaux is currently a research fellow at the ANU, and was previously research scientist at CSIRO and INRIA (France). She will be involved in the NICTA KRR program. Her research interests are in artificial intelligence, in particular in planning, model-based diagnosis, search, reasoning under uncertainty, and reasoning about action. Her favorite application areas include power/telecom/computer networks, intelligent transport systems and robotics

Seminar Convenor:

Bernhard Hengst
Tel: +61-2-9385-3988,
E-mail: bernhardh@cse.unsw.edu.au
School of Computer Science & Engineering, UNSW.