DPlan- Universal Planning and Control Rule Learning

DPlan is a domain-independent, non-linear state-based backward planner developed 1998-2000 at Technical University Berlin, Department of Computer Science, Methods of AI Group and since 2000 at University of Osnabrück, Computer Science Group (of the Department of Mathematics and Computer Science), by Ute Schmid. Modifications and extensions are supported by diploma and project students. DPlan is part of a larger system which combines universal planning, generalization-to-n, analogical reasoning and abstraction for synthesis of recursive functions (IPAL). DPlan is combined with inductive program synthesis to automatically generate domain-dependent control rules from universal plans for problems with small complexity (rocket with 3 objects, blocks-world with 3 blocks, ...). Control rules are represented as recursive functions. If control rules are learned, for this domain search can be omitted completely. The last major extension of DPlan was the inclusion of function application.

Code

DPlan is implemented in Common LISP. Programs can be interpreted with clisp or Allegro Common Lisp.

1.  [zip] DPlan-1.0 (dec 1998)

1.1   [zip] DPlan-1.1 (re-implementation, pddl format, with cond-effects, oct 2001)

2.  [zip] DPlan-2.0 (incl. function application, may 2000)

3.  DPlan uses PDDL for representing domains

4.  Graphic Output: xtree or graphlet. (called by scripts -- paths must be changed for your environment!)

4.  Work in Progress: Transformation of universal plans into initial program terms.

People

• Ute Schmid
  
• Peter Pollmanns (current diploma student): Function Application
• Janin Toussaint (collaboration 2000-2001): Application of Recursive Control Rules in Plan Construction and Transformation of Universal Plans into Initial Programs (as input to the folder which induces recursive rules).
• Marina Müller (diploma student, finished 2000): Extending DPlan to Function Application
  thesis (postscript, in German)
• Jürgen Winkler (diploma student): Domain independent platform for exploring and comparing different planning systems
  Welcome to PlaCE
• Bernhard Wolf (diploma student, finished 2000): Automatic Generation of State Sets from Domain Specifications
  [Webpage]

Papers

• Schmid, U., Müller, M., and Wysotzki, F. (2002). Integrating Function Application in State-Based Planning. 25th German Conference on Artificial Intelligence (KI2002), September 16-20, Aachen. [Postscript, 15 pages]
• Toussaint, J., Schmid, U., and Wysotzki, F. (2001). Using Recursive Control Rules in Planning. ICAI'01 Session on "Learning and Adapting in AI Planning", Las Vegas (Nevada), June 25th-28th, 2001. [Postscript, 6 pages]
• Schmid, U. and Wysotzki, F. (2000). Applying Inductive Program Synthesis to Learning Domain-Dependent Control Knowledge -- Transforming Plans into Programs. Technical Report CMU-CS-00-143, School of Computer Science, Carnegie Mellon University, Pittsburg, PA. local copy: [techrep-00.ps.gz] (126 pages)
• Schmid, U. and Wysotzki, F. (2000). Applying inductive program synthesis to macro learning. In S. Chien, S. Kambhampati, and C.A. Knoblock (Eds.), Proceedings of the AIPS 2000, Breckenridge, CO, April 2000, (pages 371-378), AAAI Press. [Postscript]
• Schmid, U. (1999). Iterative macro-operators revisited: Applying program synthesis to learning in planning. Technical Report CMU-CS-99-114, School of Computer Science, Carnegie Mellon University, Pittsburg, PA. local copy: [i-macros-1.ps.gz] (100 pages)

Projects

• "Re-Implementation of DPlan: Universal Planning With and Without predefined states, constructing dags or trees, simple forward planning, function application, typing, conditioned effects" [pdf.gz](by Peter Pollmanns und Marek Zwick)
• The following extensions of DPlan were realized in the student project from 1999 student project (in German!).
  • "Universal Planning without pre-defined initial states" [postscript] (by Michael Christmann und Stefan Rönnecke)
  • "Extension of DPlan-domain specifications to a subset of adl" [ Postscript](by Janin Toussaint, Hakan Irican and Ulrich Wagner)
  • "Empirical Comparison of Planning Systems, Automatic Generation of Sets of States" [ Postscript] (by Jan Gutsche, Andreas Stolzenberg und Bernhard Wolf)
  • "Integrating Application of recursive macros (domain dependent control rules) in plan synthesis" [ Postscript] (by Ralf Ansorg und Mischa Neumann)

Links

• Information about the last AIPS Conference (and planning competition): AIPS-2002
• Find planning groups world wide hidden beneath the states of the blocks-world problem space...

  Contact

  Email: schmid@informatik.uni-osnabrueck.de
  

  

  (last modified june/06/2002 by schmid)
  (page design by Marina Müller, May 2000)