Overview
Automated planning is a fundamental area of AI,
concerned with computing behaviors which when executed in an initial
state realize the goals and objectives of the agent. In the last 15
years, we have seen great advances in the efficiency of automated
planning techniques, as a consequence of a variety of innovations,
including advances in heuristic search for classical planning, and the
application of classical planning to non-classical planning
tasks. Nevertheless, industrial-level scalability remains a
fundamental challenge to the broad applicability of AI automated
planning techniques. This is especially notable when the space of
objects is (possibly) infinite or when there is inherent uncertainty
about the initial plan parameters.
This workshop aims to bring together researchers working on emerging
directions for addressing this challenge, including: (1) achieving
scalability through plans that include cyclic flow of control and
solve large classes of problems, (2) acquisition (through learning or
search) of domain control knowledge for reducing the cost of planning,
or otherwise structuring the space of solutions, (3) automated
composition of pre-existing control modules like software services,
and (4) synthesis of program-like structures from partial programs or
goal-specifications. Common to all of these approaches is the notion
of generalized plans, or plans that include rich control structures
that resemble programs. In addition, all of these approaches share the
fundamental problem of evaluating whether a given control structure
will be helpful in developing a scalable solution for a given class of
problem instances. While these approaches have achieved promising
results, many fundamental challenges remain regarding the synthesis,
analysis and composition of such generalized plans.
The focus of this workshop is on techniques for addressing these
challenges in particular, and more generally on scalable
representation and reasoning techniques for planning. An additional
objective is to reevaluate some of the most fundamental, traditionally
accepted notions in planning about plan structure and representation
of domain knowledge. Some of the questions motivating this workshop
are:
- How can we effectively find, represent and utilize high-level knowledge about planning domains?
- What separates planning problems from program synthesis??
- How can we effectively embed complex control structures in planning algorithms??
- What are the computational limits to the feasibility of these problems??
- Can restricted formulations of generalized planning that are practical and efficiently solvable be developed??
- How can abstraction techniques for understanding, analyzing and
reasoning about programs be utilized for generalized planning??
In addition to these key questions, we would like to additionally emphasize and encourage submissions on the following theme:
- How can we learn generalized plans and partial policies from data?
We believe a deeper integration of machine learning approaches and planning algorithms presents an exciting and novel direction for formulating and solving generalized planning.
Topics
Topics of interest to this workshop bring together research being
conducted in a range of areas, including classical planning, knowledge
engineering, partial policies and hierarchical reinforcement learning,
plan verification, and model checking.
Potential topics include but are not limited to:
- generating plans with loops
- generating parametrized plans
- program synthesis
- instantiating parametrized plans
- learning macro actions
- learning domain control knowledge
- planning with domain control knowledge (e.g., Golog, HTNs,
control rules)
- planning with partial policies
- generating robust or partial schedules
- work-flows as plans
Paper Format and Submission
We invite technical papers (up
to 8 pages), position papers
(up to 2 pages).
We invite submissions describing 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 generalized
planning. Submission of 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 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.
All papers should be typeset in the AAAI style, described at
http://www.aaai.org/Publications/Author/author.php.
Papers should be submitted via EasyChair at https://easychair.org/conferences/?conf=genplan2017.
Workshop related queries can be addressed to a common alias: genplan17
\at\ gmail.com.