Model-Lite Planning
for the Web Age Masses:
(The challenges of Planning with Incomplete and Evolving Domain Models)
Subbarao Kambhampati
Arizona State University
What is a Senior Member Paper?
• According to the conference homepage:
Has gray hair
(..and doesn’t
color it..)
• Senior Member Papers
– Seasoned experts give thoughtful critiques on
trends in the field.
“Spicy”
“No need to
code or prove”
Model-Lite Planning
for the Web Age Masses:
(The challenges of Planning with Incomplete and Evolving Domain Models)
Subbarao Kambhampati
Arizona State University
We have figured out how to scale synthesis..
Problem is Search Control!!!
 Before, planning
algorithms could
synthesize about 6
– 10 action plans in
minutes
 Significant scaleup in the last 6-7
years
 Now, we can
synthesize 100
action plans in
seconds.
Realistic encodings
of Munich airport!
The primary revolution in planning in the recent years has been
methods to scale up plan synthesis
…and we are all busy extending this
success to increasingly expressive models
• Now that we can make mince-meat of classical
problems, we turned our attention to
–
–
–
–
Temporal planning
Over-subscription planning
Hierarchical task network planning
Planning under uncertainty & partial observability
• Successively increasing model expressiveness
• Implicit in this trajectory is the assumption:
The way to get more applications is to tackle more and
more expressive domains
(Gently) Questioning the Assumption
The way to get more applications is to tackle more and more expressive domains
• There are many scenarios where domain
modeling is the biggest obstacle
– Web Service Composition
• Most services have very little formal models attached
– Workflow management
• Most workflows are provided with little information about
underlying causal models
– Learning to plan from demonstrations
• We will have to contend with incomplete and evolving domain
models..
• ..but our applications assume complete and
correct models..
Model-lite Planning
• We need (frame)work for planning that can
get by with incomplete and evolving
domain models.
– I want to convince you that there are
interesting research challenges in doing this.
• Disclaimers
– I am not arguing against model-intensive
planning
• We won’t push NASA to send a Rover up to Mars
without doing our best to get as good a model as
possible
Model-lite is in the Bible..
• Interest in model-lite planning is quite old
(but has been subverted..)
– Originally, HTN planning (a la NOAH) was
supposed to allow incomplete models of
lower-level actions..
– Originally, Case-based planning was
supposed to be a theory of slapping together
plans without knowing their full causal models
Personal motivations
• My attempts to apply planning techniques to Autonomic
Planning (ICAC 2005)
– Interested in developing automatic patching scripts (but the
difficulty was modeling..)
• My attempts to get a snap shot of public domain web
services (SIGMOD Record 2005)
– Very few of them had any formal specification (beyond some
disjointed “english descriptions”)
• My experience with data/information integration
problems (AAAI 2007 tutorial)
– Where the competing pulls from
• model-poor Information retrieval
• Model-rich data/knowledge based approaches
have lead to interest in reasoning with semi-structured (or any
sturctured) data.
Model-Lite Planning is
Planning with incomplete models
• ..“incomplete”  “not enough domain
knowledge to verify correctness/optimality”
• How incomplete is incomplete?
• Knowing no more
than I/O types?
• Missing a couple of
preconditions/effects?
Challenges in Realizing Model-Lite
Planning
1. Planning support for shallow domain
models
2. Plan creation with approximate domain
models
3. Learning to improve completeness of
domain models
Challenge: Planning Support for
Shallow Domain Models
• Provide planning support that exploits the shallow model
available
• Idea: Explore wider variety of domain knowledge that
can either be easily specified interactively or
learned/mined. E.g.
• I/O type specifications (e.g. Woogle)
• Task Dependencies (e.g. workflow specifications)
– Qn: Can these be compiled down to a common substrate?
• Types of planning support that can be provided with such
knowledge
– Critiquing plans in mixed-initiative scenarios
– Detecting incorrectness (as against verifying correctness)
Challenge: Plan Creation with
Approximate Domain Models
• Support plan creation despite missing details
in the model. The missing details may be (1)
action models (2) cost/utility models
• Example: Generate robust “line” plans in the
face of incompleteness of action description
– View model incompleteness as a form of
uncertainty (e.g. work by Amir et. al.)
• Example: Generate Diverse/Multi-option plans
in the face of incompleteness of cost model
– Our IJCAI-2007 work can be viewed as being
motivated this way..
Note: Model-lite planning aims to reduce the
modeling burden; the planning itself may actually
be harder
Challenge: Learning to Improve
Completeness of Domain Models
• In traditional “model-intensive” planning learning is
mostly motivated for speedup
– ..and it has gradually become less and less important with the
advent of fast heuristic planners
• In model-lite planning, learning (also) helps in model
acquisition and model refinement.
– Learning from a variety of sources
• Textual descriptions; plan traces; expert demonstrations
– Learning in the presence of background knowledge
• The current model serves as background knowledge for additional
refinements for learning
• Example efforts
– Much of DARPA IL program (including our LSP system); PLOW
etc.
– Stochastic Explanation-based Learning (ICAPS 2007 wkhop)
Make planning Model-lite  Make learning knowledge (model) rich
From
“Any Time” to
“Any Model”
Planning
http://rakaposhi.eas.asu.edu/model-lite
Summary
• While model-intensive planning continues to have a
place (e.g. NASA), we should also look at model-lite
planning
– Applications include workflows, web services, desktop
automation, collaborative learning/planning
• The aim is to reduce modeling burden.
– Either by reducing planning support (shallow domain models)
– or by increasing the plan creation cost (approximate domain
models)
• The challenges in each are different..
• Learning goes hand-in-hand with planning in model-lite
planning scenarios.
…It pains me to admit that a few minutes ago
I withdrew a paper from <conference> on
planning for data processing that deals with
some of these issues
From
“Any Time” to
“Any Model”
Planning
http://rakaposhi.eas.asu.edu/model-lite
Summary
• While model-intensive planning continues to have a
place (e.g. NASA), we should also look at model-lite
planning
– Applications include workflows, web services, desktop
automation, collaborative learning/planning
• The aim is to reduce modeling burden.
– Either by reducing planning support (shallow domain models)
– or by increasing the plan creation cost (approximate domain
models)
• The challenges in each are different..
• Learning goes hand-in-hand with planning in model-lite
planning scenarios.