PhD Research Talk: Planning under uncertainty: Enhancement of the FODD planner; An integer programming formulation based on Lagrangian dual decomposition

Abstract

The first part of the talk deals with the enhancement of the FODD planner. The FODD planner is a domain-independent planning program capable of handling non-deterministic actions. It was developed by Prof. Khardon's research group a few years ago. It has the First- Order Decision Diagram (FODD) data structure at its core. The performance of the planner depends on the static predicate order specified by human experts for its FODDs. We attempt to minimize this human assistance by two complementary approaches - dynamic variable reordering and a new domain model. We will discuss the ideas, implementation and experimental results in detail. The second part of the talk briefly introduces an integer programming formulation based on Lagrangian dual decomposition for stochastic planning.