Integrating transition-matrix action models into conventional planning

Determine whether and how Markov state-transition matrices that represent probabilistic action effects can be incorporated into conventional AI planning techniques, specifying a concrete methodology by which such matrices can support plan construction and reasoning within standard planning frameworks.

Background

The paper contrasts structured probabilistic representations (belief networks and conditional belief nets) with transition-matrix approaches common in stochastic modeling of actions. Although transition matrices are theoretically complete and support probabilistic projection, the authors note practical assessment burdens and inference limitations. More importantly, they explicitly state uncertainty about how such matrices could fit into conventional planning techniques, motivating their alternative structured approach.