Adaptive Constraint-Based Agents in Artificial Environments

[REFINEMENT SEARCH]   [Total-Order Planning]   [Partial-Order Planning]   [Hierarchical Planning]   [Maximal Graphs]

[ Please note: The project has been discontinued as of May 31, 2005 and is superseded by the projects of the ii Labs. There won't be further updates to these pages. ]

Hierarchical Planning

(Related publication: [PUBLink])

The approach of hierarchical planning is to first introduce coarse-grained abstract actions which are then refined to the basic actions in a stepwise manner. All possible refinements are stored in a static transformation library.

Total- and partial-order planning apply a commitment to whole actions, including all of their pre- and postconditions. In hierarchical planning, the transformation library allows us to commit to only some pre- and postconditions (by way of abstract actions) and additional ordering information. Instead of introducing a new action, a refinement step consists of applying a transformation. The propagation allows only refinements according to the transformation library. Additional ordering relations must ensure the consistency of the refinement.

Hierarchical planning is highly dependent on an appropriate transformation library. The transformation library's hierarchical structure explicitly encodes inference information and guides search toward promising plans, but because of the limited library it is often far from being complete. An interesting aspect of hierarchical planning is its application of commitments other than those to whole actions.

Examples of hierarchical planners are ABSTRIPS [PUBLink], NONLIN [PUBLink] and SHOP [PUBLink].

[REFINEMENT SEARCH]   [Total-Order Planning]   [Partial-Order Planning]   [Hierarchical Planning]   [Maximal Graphs]

For questions, comments or suggestions, please contact us.

Last update:
May 19, 2001 by Alexander Nareyek