W1. AI Planning and Learning

List of Papers

Chairs

Ugur Kuter, University of Maryland, College Park, USA
Douglas Aberdeen, NICTA, Australia
Olivier Buffet, LAAS-CNRS, France
Peter Stone, The University of Texas at Austin, USA

Preface

Great strides have been made in automated AI Planning in recent years, including very efficient planning techniques that use controlled search with domain-specific and/or domain-independent heuristics, constraint-satisfaction techniques for reasoning with time and resources, and model-checking based planning algorithms. The effectiveness of these techniques has been demonstrated in the past International Planning Competitions, and to some extent, in real-world applications such as navigation tasks, space operations, railway control, and rescue-evacuation tasks.

One challenge for most planning systems is that they require a domain expert to provide some sort of planning knowledge to the system. In many realistic planning problems, however, such planning knowledge may not be completely available; this is partly because it is very hard to compile such knowledge due to the complexities in the domains, e.g., evacuation and rescue operations, and it is partly because there is no expert to provide it, e.g., space operations. In these complex domains, a planning system that can learn such knowledge to develop ways on how to operate in the world holds great promise to be successful.

There has always been an interest in research on developing new techniques in the intersection of AI Planning and Learning. Recently, the two communities have started to move towards each other in several venues. The biggest evidence for this trend is the recent projects initiated by the major funding agencies throughout the world; e.g., the DARPA Grand Challenge, the Integrated Learning and Transfer Learning programs at DARPA, and "Dynamic Planning, Optimization, and Learning" (DPOL) project at NICTA, Australia. All of these programs can benefit greatly from research that integrates planning and learning, and to some extent have generated research in this direction.

The aim of this workshop is to bring researchers together from the AI planning and machine learning communities. The tentative topics that will be covered in the workshop include, but are not limited to:

Theory of mixing inductive and deductive approaches to planning
Learning to search (e.g., learning search heuristics)
Learning for games (e.g., learning game evaluators at end games)
Generalizing plans across similar domains
Function approximation in planning
Learning models for planning
Relational reinforcement learning and planning
Planning heuristics for exploration in reinforcement learning
Optimization based approaches to planning
Applications of planning and learning (e.g., methods applied in the past and present DARPA Grand Challenges)

Workshop website: http://www.cs.umd.edu/~ukuter/icaps07aipl/

Committee

Douglas Aberdeen, NICTA, Australia
Adi Botea, NICTA, Australia
Olivier Buffet, LAAS-CNRS, France
Alan Fern, Oregon State University, USA
Hector Geffner, Universitat Pompeu Fabra, Spain
Robert Givan, Purdue University, USA
Charles Gretton, NICTA, Australia
Guillaume Infantes, University of Maryland, College Park, USA
Subbarao Kambhampati, Arizona State University, USA
Ugur Kuter, University of Maryland, College Park, USA
Hector Munoz-Avila, Lehigh University, USA
Peter Stone, The University of Texas at Austin,USA
Sylvie Thiébaux, NICTA, Australia
Qiang Yang, HK University of Science and Technology, China
Sungwook Yoon, Arizona State University, USA
Rong Zhou, PARC, USA