Generalization in Planning

Overview

This workshop aims to build synergies across different AI communities in order to address all aspects of generalization of solutions for sequential decision making, including, but not limited to, representation of problems and solution concepts that enable efficient generalization and transfer of relevant knowledge, and algorithms for learning or synthesizing such generalized knowledge and solutions. We welcome contributions focusing on different formulations/representations for generalization, empirically validated methods, and theoretical analyses and foundations for generalization.

Humans are good at solving sequential decision making problems, generalizing from a few examples, and learning and expressing generalized knowledge that can help solve new problems. Computing such knowledge remains a long-standing open problem for Artificial Intelligence (AI). Over the last two decades, there has been remarkable progress in the performance of automated planning systems. However, real-world scalability and skill/plan generalization for complex, long-horizon tasks remains an open challenge.

We welcome contributions focusing on different formulations/representations for generalization, empirically validated methods, theoretical analyses, foundations for generalization, learning and transfer in planning.

This workshop is the sixth edition of the recurring GenPlan workshop series.

Workshop related queries can be addressed to the Organizing Committee at: genplan22\at\gmail.com.

Call for Papers

Topics of interest to this workshop bring together research being conducted not only in multiple sub-fields of AI research (including automated planning, knowledge representation, and reinforcement learning) but also in other fields of research, including formal methods and program synthesis.

We welcome submissions that address formal as well as empirical issues on topics such as:

• Formulation of generalized problems: propositional, first-order and other type of representations, and possible decomposition of such representations in terms of hierarchies, behaviors, multiple objectives, etc.
• Representation of solution structures that enable generalization and transfer.
• Learning high-level models for generalizable planning.
• Learning and synthesis approaches for computing solutions.
• Instantiation and execution of general solutions over new problem instances.
• Heuristics for plan and policy generalization.
• Learning policies and heuristics for generalized planning.
• Generation and detection of good examples for few-shot generalizable planning and learning.
• Program synthesis.
• Deriving domain control knowledge and partial policies with planning and learning.
• Generalization in environments with partial observability and/or noise.
• Generalization and transfer in reinforcement learning.

Submission Guidelines

Submissions can describe either work in progress or mature work that has already been published at other research venues and would be of interest to researchers working on generalization in planning. Previously published work in whole or in part may be in the form of a resubmission of a previous paper, or in the form of a position paper that overviews and cites a body of work. Submissions of papers being reviewed at other venues (NeurIPS, UAI, ICML, CoRL, IROS,...) are welcome since this is a non-archival venue and we will not require a transfer of copyright. If such papers are currently under blind review, please anonymize the submission.

Submissions may use either the IJCAI '22 paper format (https://www.ijcai.org/authors_kit) or the NeurIPS ’22 paper format. Camera-ready versions of accepted papers will be required in the IJCAI '22 format by the camera-ready deadline.

Two types of papers can be submitted:

• full technical papers with the length of up to 8 pages + 2 for references (9+2 pages in NeurIPS format)
• short papers with the length between 2 and 4 pages + 1 for references

Papers can be submitted via EasyChair at https://easychair.org/conferences/?conf=genplan22.

Important Dates

Invited Talks

Blai Bonet Universitat Pompeu Fabra, Spain	Feature-based Generalized Policies and Guarantees In recent years, generalized planning has become an important thread in planning and deep reinforcement learning (DRL). In the logical setting, a successful approach for generalized planning is to express general policies or strategies with rules over state features as the latter provide the necessary abstraction over classes of planning instances with different sets of grounded actions, and the rules tell which transition to take at non-goal states. Furthermore, the same type of rules can also be used to express general strategies based on subgoal-based decompositions that are guarantee to be executable in polynomial time. In this talk, we revise the main ideas underlying these approaches and address the task of establishing formal guarantees for general policies. Slides Bio: Blai Bonet is retired professor from the Computer Science Department at Universidad Simon Bolivar, Venezuela, and currently a research associate at the Universitat Pompeu Fabra, Barcelona, Spain. He received his Ph.D. degree in computer science from the University of California, Los Angeles. His research interests are in the areas of automated planning, search and knowledge representation, deep learning, and theory of computation. Blai has received several best paper awards or honorable mentions, including the 2009 and 2014 ICAPS Influential Paper Awards, and he is a co-author of the book "A Concise Introduction to Models and Methods for Automated Planning". He has served as associate editor of Artificial Intelligence and the Journal of Artificial Intelligence Research (JAIR), conference co-chair of ICAPS-12, program co-chair of AAAI-15, and has been a member of the Executive Council for ICAPS and AAAI.
Tesca Fitzgerald Yale University, USA	Abstraction in Data-Sparse Task Transfer for Interactive Robots As our expectations for robots' adaptive capacities grow, it will be increasingly important for them to reason about the novel objects, tasks, and interactions inherent to everyday life. Rather than attempt to pre-train a robot for all potential task variations it may encounter, we can develop more capable and robust robots by assuming they will inevitably encounter situations that they are initially unprepared to address. Meeting this challenge requires robots to be flexible not only to novelty, but to different forms of novelty and their varying effects on the robot's task completion. In this talk, I will focus on the relationship among (i) differences between familiar and novel task environments, (ii) the level of abstraction at which a robot's task model should be represented to enable transfer to the novel environment, and (iii) the information needed to ground the abstracted task representation in the novel environment. Slides Bio: Dr. Tesca Fitzgerald is an Assistant Professor in the Department of Computer Science at Yale University. Her research is centered around interactive robot learning, with the aim of developing robots that are adaptive, robust, and collaborative when faced with novel situations. Before joining Yale, Dr. Fitzgerald was a Postdoctoral Fellow at Carnegie Mellon University, received her PhD in Computer Science at Georgia Tech, and completed her B.Sc at Portland State University. She is an NSF Graduate Research Fellow (2014), Microsoft Graduate Women Scholar (2014), and IBM Ph.D. Fellow (2017).
Eva Onaindia Universitat Politècnica de València, Spain	Open Challenges in Learning Action Models for Planning Research in action-model learning has gained a lot of maturity since the emergence of pioneer learning systems like ARMS or SLAF. Ever since we have witnessed approaches for learning propositional and relational action models, online learning, and even systems to learning temporal and numerical action models. In this talk, we will walk through some open challenges in the field of learning action models for planning. Particularly, we will focus on investigating the type of input data and assumptions of the learning systems, the degree of observability and representativeness of the learning samples and the bias they introduce in the learning process. We will also explore the design of scalable algorithms for action model learning, including possible future directions for exploiting ML algorithms. Slides Bio: Eva Onaindia is a professor of computer science at Universitat Politècnica de València (UPV) in Spain. She has worked in several subfields of AI, particularly on the areas of Knowledge Representation and Reasoning, Automated Planning and Multi-Agent Systems. Currently, her main focus is on the integration of Planning and Learning and on the exploitation of Neuro-Symbolic approaches for decision making. She was Editor-in-Chief of the journal AICommunications, the European journal on AI, for over four years and she has served as a program co-chair of the conference ICAPS 2019.

Program

Related Workshops

This workshop builds upon GenPlan workshop series. However, new workshops on reasoning about actions include as one of their topics the generalization in planning. This is the list of most related and recent workshops:

• Workshop on Generalization in Planning at IJCAI 2021
• Workshop on Generalization in Planning at AAAI 2020
• Workshop on Reasoning about Actions and Processes: Highlights of Recent Advances at ICAPS 2019
• Workshop on Reasoning about Actions and Processes: Highlights of Recent Advances at KR 2018
• Workshop on Generalized Planning at ICAPS 2017
• Workshop on Generalized Planning at AAAI 2011
• Workshop on Generalized Planning: Macros, Loops, Domain Control at ICAPS 2009

Committees

Organizing Committee

Pulkit Verma

Arizona State University, USA

Yuqian Jiang

The University of Texas at Austin, USA

Rushang Karia

Arizona State University, USA

Jendrik Seipp

Linköping University, Sweden

Advisory Board

Blai Bonet

Universitat Pompeu Fabra

Giuseppe De Giacomo

Sapienza Università di Roma

Hector Geffner

Universitat Pompeu Fabra and ICREA

Anders Jonsson

Universitat Pompeu Fabra

Sheila McIlraith

The University of Toronto

Siddharth Srivastava

Arizona State University

Peter Stone

The University of Texas at Austin and Sony AI

Sylvie Thiébaux

Australian National University

Shlomo Zilberstein

The University of Massachussets Amherst

Program Committee

• Cameron Allen, Brown University, USA.
• Blai Bonet, Universitat Pompeu Fabra, Spain.
• Augusto B. Corrêa, Universität Basel, Switzerland.
• Giuseppe De Giacomo, Sapienza Università di Roma, Italy.
• Dominik Drexler, Linköping University, Sweden.
• Yunshu Du, Sony AI, USA.
• Salomé Eriksson, Universität Basel, Switzerland.
• Patrick Ferber, Universität Basel, Switzerland.
• Raquel Fuentetaja, Universidad Carlos III de Madrid, Spain.
• Hector Geffner, ICREA and Universitat Pompeu Fabra, Spain.
• Daniel Gnad, Linköping University, Sweden.
• Sergio Jiménez Celorrio, Universitat Politècnica de València, Spain.
• Anders Jonsson, Universitat Pompeu Fabra, Spain.
• Andres Occhipinti Liberman, Universitat Pompeu Fabra, Spain.
• Bo Liu, The University of Texas at Austin, USA.
• Reuth Mirsky, Bar Ilan University, Israel.
• Sriraam Natarajan, The University of Texas at Dallas, USA.
• Rashmeet Kaur Nayyar, Arizona State University, USA.
• Eva Onaindia, Universitat Politècnica de València, Spain.
• Hector Palacios, ServiceNow Research, Canada.
• Ron Petrick, Heriot-Watt University, United Kingdom.
• Javier Segovia-Aguas, Universitat Pompeu Fabra, Spain.
• Tom Silver, Massachusetts Institute of Technology, USA.
• Siddharth Srivastava, Arizona State University, USA.
• Simon Ståhlberg, Linköping University, Sweden.
• Roni Stern, PARC, USA and Ben Gurion University of the Negev, Israel.