Scientific context

Purpose

For this first one-day workshop on the topic of Constraints and AI Planning we take a broad view stemming from the observation that there are fundamental differences in representing and reasoning about a problem via constraints, as is common in CP and optimization, and doing so in the form of a state-transition system, as is common in AI planning, heuristic search, and dynamic programming. Therefore we hope to attract presentations and discussions on the interactions, overlaps, and differences among CP, SAT, mixed integer programming, AI planning in its many forms (e.g., classical, numeric, temporal, stochastic), heuristic search (i.e., A*-style state-based search) and dynamic programming. We are particularly interested in combinations of two or more perspectives. This workshop should interest researchers from academia and industry in the areas of constraint programming, operations research, and AI planning.

Background

While there has been a history of more than 20 years in the application of SAT, CP, and MIP to AI planning problems, there has been recent interest in a deeper cross-fertilization between the areas of constraint programming, operations research, and planning:

• A number of papers over the past five years at ICAPS (including at least two best paper winners) and planning sessions of AAAI and IJCAI (again, including at least one best paper winner) have focused on the use of techniques coming from Constraint Programming and Operations Research (e.g., LP, MIP, Logic-based Benders Decomposition) for important components of mainstream solutions to AI planning problems;
• A Dagstuhl seminar on AI planning and Operations Research (including in particular several members of the CP community) was held in February 2018;
• The 2018 editions of the ICAPS and CPAIOR international conferences are co-located.

Topics

Main areas of interest include, but are not restricted to:

• Using constraint-based techniques (e.g., CP, SAT, SMT, MIP) for solving AI planning problems including contributing to components of the AI planning solution such as heuristic evaluation.
• Theoretical/formal work comparing constraint-based and state-based representation and reasoning.
• Empirical studies comparing constraint-based and state-based solution approaches on common classes of problems.
• Extending approaches primarily developed in one area to the other (e.g., Lagrangian relaxation, Logic-based Benders decomposition, A*-style search, abstraction).
• Understanding and comparing techniques that have been developed and applied independently in both optimization/OR and AI literatures (e.g. multi-valued decision diagrams, dynamic programming vs. heuristic search, Markov Decision Processes).

Programme/Venue

Since some of the regular CP conference attendees may not be familiar with AI planning, the workshop starts with a tutorial on its fundamentals. The rest of the programme includes two invited speakers and several informal presentations and discussions based on the submitted abstracts and on the participants' interests, in keeping with the spirit of Dagstuhl seminars.

All technical presentations have a 30 minute slot including questions. Invited talks have a 45 minute slot including questions.

8h30-9h00 Registration
9h00-10h30 Session 1
Malte Helmert	AI Planning tutorial (slides)
Patrik Haslum	Invited talk: Planning with State and Trajectory Constraints (slides)
10h30-11h00 Coffee Break
11h00-12h30 Session 2
Emre Okkes Savas, Chiara Piacentini	Extending a MILP Compilation for Numeric Planning Problems to Include Control Parameters (slides)
Elad Denenberg, Amanda Coles	Expressive Plannning by Combining Forward Search and Mixed-Integer Programming (slides)
Open discussion
12h30-14h00 Lunch (provided)
14h00-15h30 Session 3
Peter Stuckey	Invited talk: Sequencing Operator Counts (slides)
Augusto B. Corrêa, Florian Pommerening, Guillem Francès	Relaxed Decision Diagrams for Delete-Free Planning (slides)
Open discussion
15h30-16h00 Coffee Break
16h00-18h00 Session 4
Stéphane Cardon	GPU-based CSP for Action Planning (slides)
Ionut Moraru, Moisés Martínez, Stefan Edelkamp	Automated Pattern Selection using MiniZinc (slides)
Guillem Francès, Hector Geffner	Constraint Propagation and Embedded CSPs in Forward-Search Planning
Open discussion
18h00 Closing

Registration

Registration for the workshops day is available from the hosting conference.

Submission

Closed

Important dates

• Abstract submission deadline: July 6, 2018
• Notification to authors: July 10, 2018
• Workshop: August 27, 2018

Organizers

• Christopher Beck, University of Toronto, jcb@mie.utoronto.ca
• Michael Cashmore, King's College London, michael.cashmore@kcl.ac.uk
• Malte Helmert, University of Basel, malte.helmert@unibas.ch
• Gilles Pesant, Polytechnique Montreal, gilles.pesant@polymtl.ca

Contact

For any questions related to the workshop, please contact one of the organizers above.