Automated Planning is the computational process of generating a course
of action for an agent to execute. As input a planner takes a
description of the world (in some formal knowledge representation), a
description of the available actions, and a description of the agent's
objective function (e.g., a goal, reward function, and / or action
costs). The planner's output may be a sequence of action (a "plan"), a
branching tree of actions (a "contingent plan") or a function from
world-state to actions (a "policy").
The computational complexity of planning depends on the expressiveness
of the languages used to represent actions and goals, but even simple
formulations are P-SPACE hard, and probabalistic representations
render the problem much harder.
The Planning group at the University of Washington has been active in
the development of planning algorithms and reprsentations since
1990. This website has more information about our research
publications, and more.