Catalogue of Artificial Intelligence Techniques
Hierarchical Task Network Planning
Aliases: Hierarchical Task Network
Keywords: compound, goal, hierarchical, network, planning, primitive, task, tasks
Author(s): George Buckenham
This is an automated planner which decomposes compound tasks into a sequence of primitive tasks. Primitive tasks can be directly executed, while compound tasks can only be decomposed into a sequence of actions, using methods. Also, there are constraints on tasks – for instance, in a journey planner, certain amounts of money must be available for a task like “ordering a ticket” to succeed. These constraints may also be in to form of requiring that certain other tasks have been performed beforehand. The planner is able to backtrack if a expansion turns out to be impossible due to constraints. The goals of the planner are specified as compound tasks which must be achieved. Goal tasks are stated in terms of conditions that must be made true.
The planner first expands each compound task into a several sub-problems. It then applies modifications to handle interactions between tasks, using critics. This is then repeated until only primitive tasks are left, at which point any conflicts (due to constraints) are resolved, and the result is returned. If the conflicts cannot be resolved, the planner fails.
- K. Erol, J. Hendler, and D. Nau, Semantics for hierarchical task-network planning Technical Report TR-94-31, UMIACS (1994).