Extending PDDL to model stochastic decision processes

Abstract

We present an extension of PDDL for modeling stochastic decision processes. Our domain description language allows the specification of actions with probabilistic effects, exogenous events, and actions and events with delayed effects. The result is a language that can be used to specify stochastic decision processes, both discrete-time and continuous-time, of varying complexity. We also propose the use of established logic formalisms, taken from the model checking community, for specifying probabilistic temporally extended goals.

Sample citation

Håkan L. S. Younes. 2003. Extending PDDL to model stochastic decision processes. In Proceedings of the ICAPS-03 Workshop on PDDL, 95–103, Trento, Italy.

Full paper (9 pages, 36 references)
Presentation (33 slides)