Program

NEW SCHEDULE.
NEW LOCATION: room BH168 (presumably Barus and Holley building)

Motivation

The automatic derivation of heuristic estimators to guide the search has become one of the main approaches in domain-independent planning. The approach works: very large problems, with many boolean variables and operators can be solved in this way, that could not be solved before, and the approach appears to scale up better than others.

Yet,

As an illustration:
  1. Consider the problem of building an ordered tower b1, ..., bn, with b1 on top, from an initial state where all blocks are on the table, except bn that is on b1. This is a trivial problem yet the extremely powerful FF planner would not solve it. Similar trivial unsolved examples can be constructed for other heuristic search planners.
  2. Or the problem of finding plans that avoid certain states; eg. states that contain certain subsets of atoms (bad states): none of the standard heuristics seems to be able to take such additional information into account.
  3. Or the more theoretical problem that in instances with quadratic length solutions, delete-relaxation heuristics are unlikely to work as delete-relaxed problems always have linear size plans.
For dealing with this and many other relevant problems, new ideas are needed. Such ideas are often difficult to publish, in the standard venues. The typical reviewing process lays a lot of emphasis on beating IPC planners. This does not leave much room for thoughtful considerations on basic properties of heuristic functions.

Workshop Objectives

In this workshop we aim to address this situation. We look for contributions that would help us understand better the ideas underlying current heuristics, their limitations, and the ways for overcoming them. Contributions do not have to show that a new heuristic or new variation of a known heuristic 'beats the competition'. While performance measured in the number of evaluated nodes, time, and solution quality remains relevant, in this workshop we seek above all crisp and meaningful ideas and understanding. The ideas have to make sense computationally, but there is no need to wrap them with many others into a planner that can be 'competitive' with state-of-the-art planners. Also, rather than being interested in the 'largest' problems that current heuristic search planners can solve, we are interested in the simplest problems that they can't actually solve well.

Workshop Format and Submission Guidelines

Papers or position statements ranging from 2 to 6 pages in AAAI format style, are invited. See http://www.aaai.org/Publications/Author/author.php for formatting instructions. The submissions should be sent by email to Carmel Domshlak (dcarmelATieDOTtechnionDOTacDOTil). The important dates are listed below.

As the workshop is not an archival conference, multiple submissions are permitted. Work that is a strict subset of that recently previously published in an archival journal should be presented as a short position statement. The workshop is planned to be a full 1-day format, but the precise format is to be decided as a function of the contributions received. We plan to have two types of presentations: long and short (15-20 and 5-10 mins), as well as 1-2 discussion panels.

Important Dates

Paper Submission Deadline June 15, 2007
Notification of Acceptance/Rejection July 13, 2007
Camera-ready Copy Due Date July 20, 2007
Workshop date September 22, 2007

Organizing and Program Committee

Hector Geffner Universidad Pompeu Fabra, Barcelona, Spain
Patrik Haslum NICTA, Canberra, Australia
Malte Helmert University Freiburg, Freiburg, Germany
Joerg Hoffmann DERI Innsbruck, Innsbruck, Austria
Vincent Vidal CRIL - IUT de Lens, Lens, France
Blai Bonet Universidad Simon Bolivar, Caracas, Venezuela
Carmel Domshlak Technion, Haifa, Israel