Augmenting Hierarchical Task Networks (HTNs) with Episodic Knowledge (e.g., cases)

Planning domain descriptions provide knowledge that indicates how an agent can act to change the environment. The importance of the first effort lies in two facts that hold simultaneously in many real-world applications. First, a complete planning domain description cannot be made available. This is due primarily because developing a complete planning domain description would require a prohibitive knowledge engineering effort. Second, cases are available in the form of concrete planning experiences performed by human experts.

Here are some publications on this topic:

• A paper describing an algorithm called SiN, which uses episodic knowledge to enhance incomplete domain descriptions. Published at the Seventeenth International Joint Conference on Artificial Intelligence (IJCAI-2001).

• A paper describing an algorithm that refines cases when an incomplete domain descriptions is available (published at ICCBR-03).

• A paper describing an algorithm that performes dependency maintenance on reused cases when an incomplete domain descriptions is available (published at ICCBR-03).

• An Ke Xu's MS Thesis on the topic of dependency maintenance on reused cases when an incomplete domain descriptions is available.

• Ian Warfield's MS Thesis on the topic of a general algorithm for dependency maintenance on top of the HTN planner SHOP. This algorithm geenralizes the one used by Ke Xu for incomplete domain descriptions.

• A paper describing DiNCAD: Case-Based Task Decomposition with Incomplete Domain Descriptions. DiNCAD uses generalized cases to extend the coverage of its knowledge base (published at AAAI-05).

• Ke Xu's dissertation on the topic of using generalized cases to enhance domain descriptions is available.

• For a recent submission, we reported on an expansion and an extension of DInCaD. For this submission we use the following files (text files) for the experiments:
  • A variant of the UMTranslog domain, originally published in: Andrews, S., Kettler, B., Erol, K., and Hendler, J., UM Translog: A Planning Domain for the Development and Benchmarking of Planning Systems. Technical Report, Dept. of Computer Science, Univ. of Maryland at College Park, 1995.
    • Here is the variant. This domain file was used to generate a case base.
    • A case base with simple generalizations of cases
    • The random problem generator (in C)
    • A problem file generated

  • An HTN variant of the process planning domain, originally published in: Muñoz-Avila, H. and Weberskirch, F., A Specification of the Domain of Process Planning: Properties, Problems, and Solutions. LSA-Report LSA-96-10E, Centre for Learning Systems and Applications, Kaiserslautern, 1996.
    • Here is the variant This domain file was used to generate a case base.
    • A case base with simple generalizations of cases
    • The random problem generator (in C)
    • A problem file generated

  • The HTN variant of the Scheduling domain used in the ICAPS International Planning Competition. This zip file contains the following:
    • schDomain-training.shp (the domain file)
    • schCases.shp (the case base with simple generalizations of cases)
    • HTNschprob.java (a random problem generator, coded in Java)
    • schprob1.shp (a problem file generated)

Last updated: Tue Sep. 25 14:49:51 EST 2007