Heuristics for Planning with Penalties and Rewards using Compiled Knowledge
Dr Blai Bonet (Universidad Simon Bolivar, Caracas, Venezuela)
NICTA KRR SEMINARDATE: 2006-12-18
TIME: 14:00:00 - 15:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
The type of heuristics considered so far in planning, only deal with simple planning models where costs are associated with actions but not with states. In this talk, we address this problem by formulating a more expressive model and heuristics where preferences in the form of penalties and rewards are associated with fluents. The heuristic is a generalization of the well-known delete-relaxation heuristic which is known to be admissible, informative but intractable. However, recent techniques of knowledge compilation allow us to compute such heuristics in practice. We'll show how suitable encodings of the planning problem in propositional theories can be compiled into d-DNNF along with interesting connections to logic programming and the stable- model semantics.
BIO:
Blai Bonet received his Ph.D. in Computer Science from the University of California, Los Angeles in 2004. In 2005 he was appointed associate professor for computer science at Universidad Simon Bolivar in Caracas, Venezuela. His research interests include automated planning, knowledge representation and reasoning, and probabilistic and graphical models.
Blai is visiting NICTA/KRR until 6 January.
