CSE574 Planning & Learning

Readings for link-analysis lectures...

2008-09-30T15:32:00.001-07:00

We are going back to search-engine technology for next class and will discuss authorities/hubs and page-rank appraoches for page importance.

The required reading is the relevant sections from

http://rakaposhi.eas.asu.edu/cse494/notes/ch2.pdf

you are also welcome to read the optional readings or Kleinberg's JACM paper and Page/Brin/Motwani's unplublished pagerank paper.

rao

Class evaluations for your (temporary) edification..

2008-05-29T23:47:00.001-07:00

Folks

It is my custom to share the college teaching evaluations with the students.

You can look at a copy of the evaluation for 574

http://rakaposhi.eas.asu.edu/tmp/574-s08.htm

(The link will be available for about a week).

Thanks to the 8 of you who took time to do the evaluation (the other four would hopefully be riddled with relentless guilt ;-)

I appreciate all the comments.

regards
Rao

The Elusive cse574 Cumulative Grade Book

2008-05-13T19:32:00.001-07:00

Folks

Here are the "grades" for the first project, the two homeworks, the final and the all important paper.

Note that for the paper, I gave the grades in four objectives:

-->"Significance" -- Is the problem being attempted significant?

--> "effort" -- How much work did the student put in to the project

--> "Presentation" -- How readable/compelling is the paper?

--> "Finish" -- Is the paper self-contained or half-finished?

The final and the domain project were graded by me. The homework 1 was graded by Menkes and homework 2 was graded by J. Benton.

The papers were all read by me as well as Will Cushing and we exchanged views on the significance etc. (This of course took the most time).

Although there is no single number for semester cumulative, I feel that the student "5297 172" certainly deserves an A+.
In keeping with my tradition, that student, if he/she so chooses, suggest me what should be the grades for others.

The grades should be on the system anytime now.. ;-)

thanks for your work. I will get in touch with some of you with more elaborate paper comments. All the graded material will be available from
my office on Friday (I am away at D.C. the next two days)

Rao
-----

Here is the link for all the semester project reports...

2008-05-07T10:14:00.000-07:00

Folks

The following link takes you to a message on the mail archive which has all the project reports sent to me

http://rakaposhi.eas.asu.edu/s08-cse574-mailarchive/msg00073.html

Rao

an old midterm and some homework solutions..

2008-05-02T19:03:00.001-07:00

Some of you wanted an example exam. Here is one from last offering (this is a midterm though)

http://rakaposhi.eas.asu.edu/cse574/1-574-midterm-soln.doc

Also, you can get solutions for the first homework here

http://rakaposhi.eas.asu.edu/cse574/hw1-f04-solns.pdf

http://rakaposhi.eas.asu.edu/cse574/hw2-f04-solns.pdf

regards

rao

your peer reviews.. all tf and no idf...

2008-05-01T15:48:00.001-07:00

I looked at your peer reviews. All I can say is that you should definitely take the classes from each other-- the lowest grade you will give seems to be 8-10 range :-)

I will have to re-scale and re-center the numbers to mine patterns from it.. There go my hopes of using just your numbers (although they might make good data for collaborative filtering question in my next fall's class).

I am slowly reading through the reports. The good news is that as of now, I like at least two of them..

rao

Please reply to this mail with your paper and (optionally) presentation as attachments

2008-04-29T19:52:00.001-07:00

thanks
Rao

bring your presentations on a memory stick..

2008-04-29T10:46:00.001-07:00

Folks

Bring your slides for today on a memory stick. We will load them on
the class computer in the beginning so the presentations can be
streamlined.

thanks
rao

Fwd: Real world blocksworld

2008-04-29T08:43:00.000-07:00

Here is some information about real world block stacking application ;1/2)

rao

---------- Forwarded message ----------
From: William Cushing <william.cushing@gmail.com>
Date: Tue, Apr 29, 2008 at 8:34 AM
Subject: Fwd: Real world blocksworld
To: "Plan-Yochan@Parichaalak. Eas. Asu. Edu"
<plan-yochan@parichaalak.eas.asu.edu>

Indeed, Hector took the picture. "Optimizing the Steel Plate Storage
Yard Crane Selection Problem", ICAPS 2007 (DC Poster). I believe a
short paper version exists in the proceedings somewhere...

picture: http://picasaweb.google.es/hectorpal/Icaps2007/photo#5116899415236524914

that group's publication list:
http://www2.imm.dtu.dk/pubdb/public/publications.php?cmd=full_view&pubtype=&section=7

A related journal article on blockstacking (no joke, that's one of the
keywords):

http://www.springerlink.com/content/f60826852715v545/

(I attached the pdf if you don't have access to springer)

-Will

---------- Forwarded message ----------
From: William Cushing <william.cushing@gmail.com>
Date: Tue, Oct 2, 2007 at 7:07 PM
Subject: Real world blocksworld
To: plan-yochan <plan-yochan@parichaalak.eas.asu.edu>

http://picasaweb.google.es/hectorpal/Icaps2007/photo#5116899415236524914

This poster is on a domain which is very blocksworld-ish. Also has
towers-of-hanois-ish in that there is no table of infinite capacity;
only stacks. However, there appear to be no stacking restrictions
(objects don't have different weights or sizes that require sorted
stacks).

The steel plates *do* have individual names which distinguish them, a
highly important technical point (in the comparison with blocksworld).

The goal state is not a stacking configuration; the goal is to
minimize movement over time (as I understand it), each day one is to
deliver certain plates to a conveyer at the far right of the
coordinate system. One has a nominal schedule when plates will
arrive and when they will be requested, allowing one to subgoal on
intermediate stacking configurations (or rather a sequence of stacking
configurations) which minimizes movements while achieving those
schedules. (and then there is some online planning for each day as
requests and supplies deviate from nominal).

So many interesting components that bar applying FF directly to the
problem...but....a big core part of the technical problem *is*
blocksworld. On ~1000 "blocks", in case anyone likes to think of
planning + scaling.

-Will

Correction for a homework 2 problem..

2008-04-28T11:32:00.001-07:00

In Question 1, where I say "Y can start at least 1 unit or at most 2
units after X" , please replace the "or" with an "and" (the "or"
constraint becomes a non-constraint since you will basically have two
constraints [-inf 2] [1 +inf] which is equivalent to [-inf +inf]

thanks
rao

Assistance or distraction?

2008-04-26T09:44:00.000-07:00

The agent is quite intelligent and his behavior is close to the optimal, he is doing well. The assistant learns the agent’s goal and behavior, and starts to act. The agent notices that the world became friendlier, but different. It is time for active exploration (which is expensive) by the agent. The exploration changes goal distribution completely; the assistant distorts the world by trying to achieve wrong goals. After awhile the assistant stops to act, the agent finishes exploration (suspecting he had a glitch); and they are back to the old world.
It seems that assistance works, if an assistant much smarter than an agent; and an agent accepting the world as it appears, without trying to learn it.

Important: Format of the class of 4/29--you will do short presentations of your project findings

2008-04-23T17:01:00.000-07:00

Folks:

For 4/29 (i.e., next Tuesday's class), I will take about 10 min to
do a "here is what we did" spiel. The rest of the class will be spent
with
each of you making a 5-min presentation about your project findings.
This will also serve as a quick summary of your report (that you will
also
bring to the class that day). Ideally, you make this presentation as
an advertisement to draw people to read your report (you will also
give me electronic--
pdf and ppt versions) of the report and talk.

I think there is a reasonable amount of diversity in the projects that
it is worth everyone hearing summaries of what others did).

thanks
Rao
-------------------
Subbarao Kambhampati
http://rakaposhi.eas.asu.edu/rao.html

(correct URL) Reading for tomorrow (as well as slides and audi...

2008-04-23T14:27:00.001-07:00

Sorry,
the URL for Minh Do's paper is

http://www2.parc.com/isl/members/minhdo/publications/2008/do.pdf

rao

---------- Forwarded message ----------
From: Subbarao Kambhampati <subbarao2z2@gmail.com>
Date: Wed, Apr 23, 2008 at 2:26 PM
Subject: [CSE574 Planning & Learning] Reading for tomorrow (as well as
slides and audi...
To: subbarao2z2@gmail.com

Folks

For Minh Do's guest lecture tomorrow, you might want to read the
following paper

http://www.aaai.org/Conferences/AAAI/2008/aaai08nectar.php

(it is
only 5 pages!)

(Additional related papers are on his homepage at
http://www2.parc.com/isl/members/minhdo/ )

----------

Also, for those of you who missed Biplav Srivastava's talk on web
service composition today, his slides
and audio are up on the class notes page http://rakaposhi.eas.asu.edu/cse574

see you tomorrow.

Rao

--
Posted By Subbarao Kambhampati to CSE574 Planning & Learning at
4/23/2008 02:26:00 PM

Reading for tomorrow (as well as slides and audio from today's talk)

2008-04-23T14:26:00.001-07:00

Folks

For Minh Do's guest lecture tomorrow, you might want to read the
following paper

http://www.aaai.org/Conferences/AAAI/2008/aaai08nectar.php

(it is
only 5 pages!)

(Additional related papers are on his homepage at
http://www2.parc.com/isl/members/minhdo/ )

----------

Also, for those of you who missed Biplav Srivastava's talk on web
service composition today, his slides
and audio are up on the class notes page http://rakaposhi.eas.asu.edu/cse574

see you tomorrow.

Rao

Fwd: teaching evaluations

2008-04-23T10:14:00.001-07:00

You are so encouraged! As usual, in addition to the numbers, it will
be helpful to have
written comments on what worked and what didn't.

thanks
rao

---------- Forwarded message ----------
From: James Collofello <JAMES.COLLOFELLO@asu.edu>
Date: Wed, Apr 23, 2008 at 9:59 AM
Subject: teaching evaluations
To: "DL.WG.CEAS.Faculty" <DL.WG.CEAS.Faculty@mainex1.asu.edu>
Cc: Ann Zell <ann.zell@asu.edu>

Colleagues,

The Spring 2008 teaching evaluations became available to students on
Mon 4/21, around 10:30 a.m. and will close at Wed 4/30 (reading day)
at 12:00 midnight. Students will be able to access the evaluation
tool at: https://fultonapps.asu.edu/eval

Please encourage your students to complete the evaluations, especially
the students who enjoy your class, or they will face several nagging
email requests.

jim

James S. Collofello

Associate Dean for Academic and Student Affairs

Ira A. Fulton School of Engineering

Refreshed the plan recognition slides from the first plan recognition lecture..

2008-04-23T09:22:00.001-07:00

Folks

FYI, I revised and expanded the plan recognition slides from the
first lecture based on the actual discussion in the class.

rao

Reminder: Talk Today 11AM on "The Myth and Reality of Web Services Composition" by Biplav Srivastava of IBM research 4/23 11AM BY 510

2008-04-23T06:59:00.001-07:00

Title:
The Myth and Reality of Web Services Composition

Speaker:
Dr. Biplav Srivastava
IBM Research

Time: 11AM--12AM; Wednesday April 23rd; BY 510

Abstract:

Divide-and-conquer or working with complex systems from their
basic building blocks is one of the basic tenets of modern engineering.
While its applicability to Information Technology has always been
felt – example Object Oriented Methodology, its success has been
limited. There is a resurgence of interest in componentization of
IT systems and services through focus on Service Oriented Architecture,
and Web Services as its most popular form. Consequently, Web Services
has received wide attention in both academia and IT industry over
the past 5-7 years. The attractiveness of this technology lies in
the fact that the specifications of the building
blocks (i.e., services) are openly available in a registry and
so are the building blocks themselves. So, the promise is that a user
can build (or modify) an application by composing (or re-composing)
components whose specification it discovers from the registry
and whose capabilities it can access whenever needed. Depending on
what is defined as a service, web services composition can enable
many IT issues -- Mashups, Asset Reuse, Business-to-IT alignment,
Business-to-Business and Enterprise Application Integration, ...

In this talk, we will look at where the hardness of automatically
composing web services comes from in practice and how traditional
Computer Science techniques (notably planning) have fared. While
the original myth was that composition would be hard, in reality,
most composition scenarios did not demand scalability of the
top-of-the-line planning algorithms. However, what has turned
out to be harder than composition is how to set up the composition
problem as a traditional Computer Science (notably planning) problem.
Two trends are emerging to address this: the composition problem is
often cast as a plan reuse and modification problem in the context of
richer domain models (e.g., Industry Business Processes), and new
composition/ planning paradigms like model-lite planning which are
resilient to impoverished domain models.

-----------

About Biplav:

Dr. Biplav Srivastava is a Research Staff Member at IBM Research since
February 2001. Though based at IBM's India Research Laboratory, Biplav
is on assignment to IBM's T.J.Watson Research Center in Hawthorne, NY, USA.
Biplav's research interests are in planning, scheduling, policies,
learning and information management, and their practical applications
in services -- infrastructure and software (web services), semantic web,
autonomic computing and societal domains. Prior to IBM Research, he
was Core Technology Architect at an erstwhile Silicon Valley
start-up, Bodha, eventually acquired by SAP (2000-2001; process integration),
Staff Software Engineer at VLSI/ Philips Semiconductors (1996-2000;
electornic design automation) and Assistant System Analyst at TCS,
India (1993-1994).

Aiming to minimize the expected fallout from the unexpected hanging... ;-)

2008-04-22T17:57:00.001-07:00

I will accept one page (printed) on your review/insight on the
decision theoretic assistance paper
at the *beginning* of thursday's class. Credit for saying things that
I didn't particularly bring out.

cheers
Rao

Talk on "The Myth and Reality of Web Services Composition" by Biplav Srivastava of IBM research 4/23 11AM BY 510

2008-04-22T06:18:00.001-07:00

[CSE574 Folks:

This is the first of the two application talks planned for this week.
The second one will be in class on Thursday. Please make every effort
to attend. -Rao]

Title:
The Myth and Reality of Web Services Composition

Speaker:
Dr. Biplav Srivastava
IBM Research

Time: 11AM--12AM; Wednesday April 23rd; BY 510

Abstract:

-----------

About Biplav:

talk on Online Continual Planning to Control Modular Production Printer (Minh Do; PARC Labs) 4/24 3:15pm BY 190

2008-04-22T06:16:00.000-07:00

Title: Online Continual Planning to control Modular Production Printer

Speaker: Minh B. Do, (Xerox) PARC Labs, Palo Alto

Date/Time: Thursday 4/.24 3:15pm BY 190

Abstract: This talk summarizes the recent work at the Embedded
Reasoning System at PARC on applying automated planning techniques to
the control of modular production printing equipments. These
reconfigurable printers radically change the traditional design by
using simpler, interchangeable, but smarter components. Like many
other real-world applications, such as mobile robotics, this complex
domain requires real-time autonomous decision-making and robust
continual operation. To our knowledge, this work represents the first
successful industrial application of embedded domain-independent
temporal planning. Main challenges of applying automated planning
technology in this domain include compositional modeling, on-line
planning and exception handling, real-time planner control, and the
interaction with low-level controller. At the heart of our system is
an on-line algorithm that combines techniques from state-space
planning and partial-order scheduling. For example, our
planning-graph-based planning heuristic takes resource contention into
account when estimating makespan remaining. We suggest that this
general architecture may prove useful as more intelligent systems
operate in continual, online settings. Our system has been used to
drive several commercial prototypes and numerous hypothetical (but
realistic) printer configurations. When compared with
competition-winning state-of-the-art off-line planners in this domain,
our system is hundreds of times faster and often finds much better
quality plans. At the end of the talk, I will also discuss current
extensions of our current planning framework to other online planning
domains that share similar characteristics and also to objective
functions beyond the default maximization for machine productivity.

Bio: Minh Do is a Research Staff in the Embedded Reasoning Area at the
Palo Alto Research Center (formerly Xerox PARC). He graduated from the
Yochan planning group at Arizona State University in 2004 and has been
working on transferring his knowledge in offline domain-independent
metric temporal planning into fast online continual planning
applications. Besides temporal and online planning, Minh Do has
worked on other planning topics such as over-subscription planning,
planning as CSP/ILP/SAT, and integrating planning and diagnosis. He
has published a few dozens papers, filed several patents on automated
planning and co-authored the ICAPS best application paper award on
planning for high-speed modular printer control. This year, he is
co-chairing the deterministic track of the 6th International Planning
Competition.

Re: [CSE574 Planning & Learning] readings for next class

2008-04-21T14:01:00.001-07:00

should work now

rao

On Mon, Apr 21, 2008 at 1:50 PM, Aishwarya Sivaraman <asivaram@asu.edu> wrote:
> Dr. Rao,
>
> I am unable to open the longer versions' link. I get a 404 error.
>
> http://rakaposhi.eas.asu.edu/cse574/notes/asst.pdf
>
>
> Regards,
> Aishwarya
>
>
> On Mon, Apr 21, 2008 at 1:23 PM, Subbarao Kambhampati
> <SUBBARAO.KAMBHAMPATI@asu.edu> wrote:
>
> > For the decision theoretic assistant paper, if you find the short one
> > too succinct, you might
> > consider the longer version here
> >
> > http://rakaposhi.eas.asu.edu/cse574/notes/asst.pdf
> >
> > rao
> >
> >
> > On Sun, Apr 20, 2008 at 9:32 PM, Subbarao Kambhampati
> > <subbarao2z2@gmail.com> wrote:
> > > Folks
> > > Here are two readings for next class:
> > >
> > > Primary: http://www.cs.orst.edu/%7Eafern/papers/ijcai07-assistant.pdf
> > >
> > >
> > > Secondary:
> > >
> > > http://www.cs.rochester.edu/~kautz/papers/gps-tracking.pdf
> > >
> > >
> > > Rao
> > >
> > > --
> > > Posted By Subbarao Kambhampati to CSE574 Planning & Learning at
> 4/20/2008
> > > 09:32:00 PM
> > >
> > >
> > >
> > >
> > >
> > >
> >
>
>

Re: [CSE574 Planning & Learning] readings for next class

2008-04-21T13:23:00.001-07:00

For the decision theoretic assistant paper, if you find the short one
too succinct, you might
consider the longer version here

http://rakaposhi.eas.asu.edu/cse574/notes/asst.pdf

rao

On Sun, Apr 20, 2008 at 9:32 PM, Subbarao Kambhampati
<subbarao2z2@gmail.com> wrote:
> Folks
> Here are two readings for next class:
>
> Primary: http://www.cs.orst.edu/%7Eafern/papers/ijcai07-assistant.pdf
>
>
> Secondary:
>
> http://www.cs.rochester.edu/~kautz/papers/gps-tracking.pdf
>
>
> Rao
>
> --
> Posted By Subbarao Kambhampati to CSE574 Planning & Learning at 4/20/2008
> 09:32:00 PM
>
>
>
>
>
>

readings for next class

2008-04-20T21:32:00.001-07:00

Folks
Here are two readings for next class:

Primary: http://www.cs.orst.edu/%7Eafern/papers/ijcai07-assistant.pdf

Secondary:

http://www.cs.rochester.edu/~kautz/papers/gps-tracking.pdf

Rao

Link to the goal graph based plan recognition discussed in today's class..

2008-04-17T20:28:00.001-07:00

is here
http://www.cs.cmu.edu/afs/cs/project/jair/pub/volume15/hong01a.pdf

Rao

ps: Here is a paper talking about minimizing a possibly non-minimal plan. It shows the NP-completeness of
minimization and talks about polynomial sound-but-incomplete minimization algorithms

http://citeseer.ist.psu.edu/fink92spectrum.html

an efficient temporal planner for handling required concurrency..

2008-04-16T09:15:00.001-07:00

Here is a planner to appear in AAAI that realizes Tempo (described in the temporal planning paper you read) and
shows that it is efficient (recall that VHPOP like planners could already solve RC problems--while
DEP planners couldn't. Tempo was supposed to be a way of combining reachability heuristics with non-DEP
search space. this planner--Crikey3--realizes Tempo by making several interesting improvements to
SAPA's relaxed plan graph heuristics).

FYI
rao

http://www.cis.strath.ac.uk/cis/research/publications/papers/strath_cis_publication_2248.pdf

Reference on McLug

2008-04-15T20:59:00.001-07:00

I discussed McLug at great speed towards the end of the class. Here are ways you can get more information on it:

You can look at the relevant section in the planning graph heuristics survey paper (that you had already printed):

http://rakaposhi.eas.asu.edu/pgSurvey.pdf

or you can read an entire paper describing it.

http://rakaposhi.eas.asu.edu/dan-aij.pdf (journal version)

http://rakaposhi.eas.asu.edu/ICAPS0602BryceD.pdf (older conference version)

Rao

ps: Check out the CFP for a 2008 ICAPS workshop on planning under uncertainty, to get an overview of some of the
hot/provocative issues in the area:

http://www.ai.sri.com/~bryce/ICAPS08-workshop.html

Reading for next class

2008-04-15T19:24:00.001-07:00

I am trying to find an easy survey for plan recognition--but as of now the best I could manage is the background chapter of Nate Blaylock's thesis.
It is 16 pages double spaced, and is available at

http://rakaposhi.eas.asu.edu/cse574/blaylock-pr.pdf

So unless I find something better, this is your reading for the next class.

Rao

(IMPORTANT) Schedule for the rest of the semester etc.

2008-04-13T14:17:00.000-07:00

Folks:

Here is a heads up on the rest of the semester.

===========

Class:

We will complete the treatment of FOMDPs on Tuesday (make sure you do the readings).

For the next topic--which will be discussed on Thursday and Tuesday 4/22, I am trying to decide between POMDPs (partially observable MDPs) and plan recognition.

(If you have any preference let me know).

In the week of 4/22, we will likely have 2 external speakers who will talk on applications of planning.

On 4/23 (Wednesday), Biplav Srivastava from IBM will speak on planning for web service composition

On 4/24 (Thursday)--during the class time--Minh Do will speak about applications of planning technology at Xerox PARC

(where they use it to plan paper paths through copiers!).

4/29 will be recap and summary. You will be doing an interactive review.

=============

Assessment:

1. The class project report/paper will be due on the last day of classes (4/29).You should give yourself at least 10 days to write the report. Incoherent and badly organized reports will not be good for your grade.

2. The second homework also will be due on 4/29

3. The final exam for the class will be on the normal final exam day--which is May 6th (2:40--4:30pm)

=====================

cheers

Rao

Readings for Tuesday's class

2008-04-10T17:49:00.001-07:00

A paper on the probabilistic planning competition (the first one). Describes PPDDL standard as well as
thumbnail sketches of competitors

http://www.jair.org/media/1880/live-1880-2554-jair.pdf

A paper on solving stochastic planning problems using determinizations
(it describes the algorithm that seems to work best on the current benchmarks)

http://rakaposhi.eas.asu.edu/tmp/hh.pdf

(note that this is a paper to appear in AAAI 2008. The proceedings version will be available on Tuesday (which is the deadline
for sending it :-)

You may also, optiionally, read the following critique of probabilistic planning competition:

http://www2.parc.com/isl/members/minhdo/icaps07_ws/papers/ICAPS06LittleI.pdf

Rao

Homework 2

2008-04-04T20:13:00.000-07:00

Folks

I put up some questions for homework 2. I expect to put more questions--on temporal planning, MDPs etc soon.

This homework will be due by the end of the semester. So you may want to work on it when you have time rather than

all at once...

I will send mails when I add more questions

rao

Cooperative planning using multiple agents

2008-04-03T17:20:00.000-07:00

I haven't seen many comments on the blog for awhile, so here's something for discussion.

The topic of cooperative planning between multiple agents came up in another class I attend. No one in that class had heard of much research on this front. The questions are, what could be done, what has been done, and what interesting problems are there in this field?

I suggested at the time that if each agent was an executioner of the plan then these agents could be viewed as resources, and one among them could act as a scheduler assigning parts of the plan to each agent. The plan would be divied up and each agent given a sub plan that when all executed together would complete the larger plan. This is sort of cheating though because all of the planning is then being done by one agent, its just the doing that is cooperated. This was really all that I knew of that had been done in planning that could contribute to this topic.

The next bit was applying game theory. Do the agents assume the other agents are cooporative? How is a goal established? How does each agent plan around the possibility that other agents do not complete their part? How much communication and sensing do each of these agents have during planning and execution? In the case of non perfect communication could these agents adapt to unexpected problems during execution and adjust the plan?

If you know of successful research along this topic post a link to it. Otherwise feel free to post thoughts on any part of this.

Plan (and readings) for next Tuesday's class (4/8)

2008-04-03T09:26:00.000-07:00

Folks:

As I mentioned, I will be away on travel next Tuesday (4/8). Sungwook Yoon will
lecture on Reinforcement Learning--which will be a natural continuation of (FO)MDPs that we are discussing
this week. RL involves interleaving learning, planning and execution.

The reading for 4/8 is the chapter 21 in Russell and Norvig.

regards
Rao

ps: Regarding the idea of having an assessment exam on 4/8--the feedback I got was in favor of exam at the end of the semester.

Readings for Tomorrow--LAO* and LRTDP

2008-04-02T06:56:00.001-07:00

Part of what we will do tomorrow is to consider optimal policy
construction algorithms for
FOMDPs that make it look more like heuristic search that we have seen
for the rest of the semester.

In particular, we saw that classical planning can be seen as A*
search, and belief-space planning can be seen as
AO* search. Typical AO* search algorithms work on acyclic graphs (note
that AO* can be seen as a
problem decomposition framework, and cycles imply that you are
reducing a problem indirectly to itself).
The LAO* paper below shows that FOMDP policy construction can be seen
as AO* search on cyclic
graphs.

LAO*
http://www.cse.msstate.edu/~hansen/papers/laostar.pdf

Another idea for viewing value function computation is in terms of
fixed-depth expansion under a node
(as in game trees--in fact, in 471, I motivated game trees in terms of
RTDPs). The LRTDP algorithm
improves a bit on RTDP

LRTDP
http://ftp.cs.ucla.edu/pub/stat_ser/R319.pdf

rao

Reminder: Reading for next class: Decision Theoretic Planning

2008-03-31T16:15:00.001-07:00

Folks:
A reminder about the reading for tomorrow's reading.
Please note that about half the students who took 471 with me have already
had some introduction to MDPs. So, the discussion is likely to be faster paced.
It is thus imperative that those who haven't had exposure to MDPs read
the material
once before coming to class.

thanks
rao

---------- Forwarded message ----------
From: Subbarao Kambhampati <rao@asu.edu>
Date: Thu, Mar 27, 2008 at 6:12 PM
Subject: Reading for next class: Decision Theoretic Planning
To: Rao Kambhampati <rao@asu.edu>

Here is the paper for the next week.

http://www.cs.cmu.edu/afs/cs/project/jair/pub/volume11/boutilier99a.pdf

You should read at least upto page 35 for next week.

Rao

Fwd: [CSE574 Planning & Learning] Reminder: Seminar today by Dana Nau (11:30AM BY2...

2008-03-31T11:19:00.000-07:00

---------- Forwarded message ----------
From: Subbarao Kambhampati <subbarao2z2@gmail.com>
Date: Mon, Mar 31, 2008 at 6:42 AM
Subject: [CSE574 Planning & Learning] Reminder: Seminar today by Dana
Nau (11:30AM BY2...
To: subbarao2z2@gmail.com

Planning for Interactions among Autonomous Agents

Dana S. Nau
Department of Computer Science
University of Maryland
College Park, MD

Date: Monday, March 31, 2008
Time: 11:30 AM - 12:45 PM
Place: Brickyard Building, Room 210

This talk will focus on ways to plan an autonomous agent's
interactions with other autonomous agents. Sometimes it is feasible to
model the other agents' possible actions as nondeterministic outcomes
of our agent's own actions. For this case, we can plan how to achieve
a desired goal using symbolic model checking, HTN planning, or a
combination of the two. Sometimes it may not be feasible to generate a
plan or policy that goes all the way to a goal, either because the
search space is too large or the goal is ambiguously defined. For such
cases, it can work well to interleave planning and execution if we
have a good predictive model of how the other agents are likely to
behave. The talk will present theoretical foundations and algorithms
for the above situations, and experimental results on the
Hunter-and-Prey domain, the Iterated Prisoner's Dilemma with Noise,
and other multi-agent planning domains.

Biography

Dana Nau is a Professor of both Computer Science and Systems Research
at the University of Maryland, and is co-director of the university's
Laboratory for Computational Cultural Dynamics.He has more than 300
refereed technical publications on automated planning, search
algorithms, game theory, and other topics.He has received an NSF PYI
award, several best-paper awards, and several prizes for the
performance of software systems.He is a Fellow of the Association for
the Advancement of Artificial Intelligence (AAAI).

==============

--
Posted By Subbarao Kambhampati to CSE574 Planning & Learning at
3/31/2008 06:42:00 AM

Reminder: Seminar today by Dana Nau (11:30AM BY210) on "Planning for Interactions among Autonomous Agents"

2008-03-31T06:42:00.001-07:00

Planning for Interactions among Autonomous Agents

Dana S. Nau
Department of Computer Science
University of Maryland
College Park, MD

Date: Monday, March 31, 2008
Time: 11:30 AM - 12:45 PM
Place: Brickyard Building, Room 210

Biography

==============

Need for assessment in the class??--feedback requested (can be on the blog)

2008-03-28T06:23:00.001-07:00

Folks

Now that I have an idea of what semester projects are being attempted
by the various students, I am beginning to wonder
what the grade should be based on in the event that the semester
project doesn't end up all that well.

Now, in real world, if your project doesn't work, your paper will just
be rejected and that is that. However, since this is a course
and you need a letter grade, we need some basis for that. [There is
the option of just giving everybody the same high grade--but I am
afraid that might adversely affect my popularity...]

As of now, we had one homework, and one mini-project. I definitely
plan to release a second homework to cover everything since the
first homework.

In addition to that, I am thinking that there should be some sort of
"breadth of understanding" assessment. We can do this assessment
either around now (specifically, there is a good chance I will miss
the class of April 7th-- and that is always a goodtime to have an
exam), or
at the end of the semester during the finals week.

My assumption is that if someone does a great semester project, that
trumps over everything else. However, if the semester project doesn't
work out all that well--these other assessment mechanisms can help in
giving a grade.

I solicit your opinion(s) on this (either on the blog or by anonymous
email to http://rakaposhi.eas.asu.edu/cgi-bin/mail?rao )

thanks
Rao

Over-sensing and LCW actions.. (as well as planning with sensing as internet planning)

2008-03-28T06:02:00.001-07:00

1. We talked at length about how over-sensing during execution can
delay things and lead to
decidedly non-intelligent behavior (e.g. Sphexishness).

One relevant issue here is that if you *know* the state of the world
in some aspect, and you know that
you haven't done anything to change it, then there should be no reason
for you to check it again.

Assuming a single-agent world, if you start with complete state and do
deterministic actions, then you never have
to look (as we discussed). Even if we start with incomplete initial
state, we may know "everything" about *some aspects*
of the world. As long as the ensuing actions--of ourselves--don't
modify the completeness of that knowledge, we don't
need to look *for that aspect* of the world.

It all starts to look like managing "closed world" assumptions. In the clasical
planning, we know init state, and so we can start with closed-world
assumption. No actions modify that closed world assumption.
In general belief-space planning, we don't have full closed world
assumption, but may have *partial* closed world assumption. As
we do actions, we may lose or acquire closed world assumption. In a
desktop (or unix) world, for example, we may start knowing
names of the files in the current directly as well as the sizes of all
the files in the directory. After we run a latex command, we still
know
the names of all files in the directory (even though latex makes new
files-- we know what they will be -- .aux, .bbl etc). We however no
longer
know the sizes of all files (since the sizes of .aux and .bbl files
generated will depend on the files you latexed in a complex way and
you can't
model it a priori). So, we lose closed world knowledge of the file
sizes. If we need that, we will have to do an "ls -s" action (a
sensing action).
If we just happen to do "rm *" action in that directory, we again get
full knowledge on both files and sizes of the directory.

In the paper below
http://www.cs.washington.edu/homes/etzioni/papers/xii-aaai94.pdf
Golden et. al. formalize this notion of starting with and tracking
local closedworld assumptions (LCWs) .
Their main contribution is to note not only the normal effects of the
actions, but also their meta-effects on closed world assumptions
(e.g. see the latexing and rm'ing actions above). Neat paper to read.

=========================================

2. When discussing progression planning in the presence of sensing
actions, I pointed out that there are two non-deterministic
branches: one which picks a causative action to execute and the other
which picks a sensing action to execute.
I also mentioned that if you always pick the causative action branch,
you get "conformant" or "no-sensing" plans (if you succeed).

A related question is what happens if you always pick only sensing
action branch? You get a pure sensing plan.
We can see a use for "pure causative plan" (conformant plan)--an agent
which has no sensors has to deal just with those.
Of what use can pure sensing plans be?

Well--if all you can do is sense some database, then your plans will
be just pure sensing actions. In particular, plans whose main purpose
is to
gather information can be thought of in terms of pure (or almost
entirely) sensing plans.

When you do planning on the web--for example--most often, your actions
involve sensing (look at this database, take a value from there and
plug it into a sensing query for this other database etc.) that leave
the world as it is, and only modify *your knowledge* of it.
Of course, you can also sometimes have causative actions (e..g.
updates--credit card
transactions etc) that modify the state of some database, and not just
the state of your knowledge.

So, not surprisingly, planning for information gathering involves
mostly things of pure sensing actions. The good part about sensing
actions is
that there are never any negative interactions among them (your brain
doesn't explode because you learned knowledge in the wrong order ;-).
So, for sensing planning the big issue is not so much about subgoal
interactions, but rather about reducing sensing.

Not surprisingly, the LCW stuff discussed above--wind up being relevant.

See http://rakaposhi.eas.asu.edu/ijcai-ig.pdf
which talks about how LCW information can be used to reduce the number
of information sources
that the agent has to sense to get all answers for a query.

(The paper http://rakaposhi.eas.asu.edu/ig-tr.pdf
provides a somewhat dated tutorial introduction to planning for
information gathering).

Rao

Reading for next class: Decision Theoretic Planning

2008-03-27T18:12:00.001-07:00

Here is the paper for the next week.

http://www.cs.cmu.edu/afs/cs/project/jair/pub/volume11/boutilier99a.pdf

You should read at least upto page 35 for next week.

Rao

Re: A short little paper for next class reading..

2008-03-25T18:06:00.001-07:00

You can also get a much shorter overview by reading pages 73-76 of

http://rakaposhi.eas.asu.edu/pgSurvey.pdf

(the AI Magazine tutorial paper on PG heuristics that you already
printed and read a part of
back when we were doing classical planning heuristics)

Rao

On Tue, Mar 25, 2008 at 5:44 PM, Subbarao Kambhampati <rao@asu.edu> wrote:
>
> http://rakaposhi.eas.asu.edu/dan-jair-pond.pdf
>
> (As an added bonus, section 2 provides an overview of the progression
> and regression with and without sensing actions in belief space)
>
> Rao
>

A short little paper for next class reading..

2008-03-25T17:44:00.001-07:00

http://rakaposhi.eas.asu.edu/dan-jair-pond.pdf

(As an added bonus, section 2 provides an overview of the progression
and regression with and without sensing actions in belief space)

Rao

Added several slides to yesterday's lecture

2008-03-21T10:50:00.000-07:00

I added several slides to yesterday's lecture to better reflect the
discussion in the class.

rao

The references for the complexity results on temporal planning and planning with partial observability/non-determinism

2008-03-20T22:32:00.001-07:00

Here are the papers

Complexity under partial observability/non-determinism

http://users.rsise.anu.edu.au/~jussi/RintanenICAPS04.pdf

Complexity of temporal planning

http://users.rsise.anu.edu.au/~jussi/Rintanen07icaps.pdf

Rao

Why is this still a problem?

2008-03-18T19:41:00.001-07:00

What is the main problem in doing an analysis on the ruleset before the planner gets to work to determine which constraints are reasonable to disregard during the planning process? Does it make programming the ruleset too hard up front? The number of available resources has to be defined anyway right? So the only extra work is providing an estimation of the usage of each resource or an estimation of its availability. If there's 10 taxi in the city you have an impression that you won't be able to find one because more than 10 people use a taxi in an your travel window to get from place to place. In a more cognitively plausible sense we just have an impression that taxi's will be available from past experience.

If it is unreasonable to expect someone to tell us this availability information directly there is still easy work that could be done. There are easy cases that could be taken care of up front. In the blocks world example, if the world has four arms, and you only have 3 blocks, its evident enough that the number of arms aren't a restriction and so this responsibility can be passed on to the scheduler.

In a more realistic project planning example you could create a graphplan for a few levels deep and count the number of actions that are done in parallel on each level and compare this to resource availability to determine which constraints are unnecessary. Using this information you could change the domain on the planner side to be simpler and then proceed using whatever method you wish from there.

So, why is this concept at the forefront of research? There seem to be obvious places to make initial progress, I would think it would be standard in implementations by now.

Reading for Thursday's class (belief states...)

2008-03-18T18:03:00.001-07:00

This would be a good reading for the next class

http://rakaposhi.eas.asu.edu/cse574/aips00-incomplete.pdf

Rao

Is CSP scheduling better than IP scheduling?

2008-03-10T21:27:00.000-07:00

Do I understand correctly that the slides say that CSP scheduling (based on PCP) performs better than IP based scheduling? That sounds strange: because there are mature IP-solvers and there are straightforward IP encodings (that are not part of AI, I believe) for job-shop scheduling problems. More that that, PCP does not manage well multi-capacity resources (which are widespread); and that kind of condition should not matter to IP.
Oleg Bakun

Planning and Scheduling

2008-03-08T05:57:00.000-08:00

Some most important discussions in the class were:

Planning and scheduling are separate problems as such but go hand-in-hand. A scheduled plan is one which has the plan and a schedule for that plan with time and resource constraints.

Scheduling can be:
-called as a special case of planning with resource constraints
-called as an allocation problem that allocates resources to time points and resources

A planner solves the problem and a plan serves as an input to the scheduler. This input is similar to a disjunctive temporal network which can have either unary or multi capacity resources which are nothing but unary or n-ary mutual exclusions.

Even when the resources are available before or while planning, we would want to go in a hierarchical fashion, ignore the resources and produce a plan and leave the rest to a scheduler for computational reasons.

Planning and Scheduling can be considered as two ends of a spectrum with maximum and minimum level of choices respectively. There are a number of interesting problems in-between these two points in the spectrum.

A scheduling problem can be modeled as a CSP.
Start point representation and PCP representations for CSP were discussed.
PCPs do perform well as it converts the the disjunctive temporal constraint problem into simple temporal problems that are really much more feasible when compared to the start point representation.
One problem with PCP representation is they do not scale very well for multi resource constraints.

Re: CSE574 grades

2008-03-06T16:35:00.001-08:00

I hope to give you the projects and homeworks and term paper proposals back
by the time you get back from Spring break.

Regarding the relative weight of the assignments--I am not too sure
yet. In general
I think about [30- 40%] for homeworks etc, [10-20%] for participation
and [40% -50%] for semester project.

By the way, I should tell you that one idea I am toying with is to do
an oral midterm
(Y'all get to come for half-an-hour each and the teaching czars and I
will "interview" you
on what happened until now).

Rao

On Tue, Mar 4, 2008 at 7:59 PM, Oleg Bakun <Oleg.Bakun@asu.edu> wrote:
> Hello,
> When grades and their averages for HW1 and Project 1
> will be published?
> What is relative weight of each assignment?
>
> Oleg Bakun
>
>
>
>
> ____________________________________________________________________________________
> Looking for last minute shopping deals?
> Find them fast with Yahoo! Search.

http://tools.search.yahoo.com/newsearch/category.php?category=shopping
>

Slack-Based Heuristics Scheduling Paper

2008-03-06T13:09:00.000-08:00

The slack-based heuristics for constraint satisfaction scheduling paper visits a number of points that we have seen through the semester, although it doesn't call them what we know them as. Here are a couple of them:

1. In the introduction, 2 different methods of constraint-based scheduling are mentioned: that of posting start times, and the method which the authors use, of posting sufficient additional sequencing constraints between pairs of operations contending for the same resource. It should be really clear that (in planning lingo) this is in essence totally-ordered plans vs. partially-ordered plans: as the authors point out, the solutions generated by the latter method "... typically represent a *set* of feasible schedules" (emphasis added). These are but partially-ordered plans; any topological sort of them will yield a total-order that is executable.

2. This second point probably has a lot to do with the fact that the method proposed deals with constraints as a CSP would: the initial configuration of variable and value ordering heuristics as defined by the authors is something we have already seen in CSP.

Min-Slack: This is similar to the Most-Constrained Variable (MCV) selection in CSP.

Max-Slack: This is similar to the Least-Constraining Value selection in CSP.

kartik

Added slides on quotienting and candidate set semantics for temporal planning

2008-03-06T07:22:00.000-08:00

I added slides to last classe's lecture notes to reflect the
whiteboard discussion we had about
quotienting and lifted spaces and candidate set semantics for lifted
planning (this was done in
response to a question from Tuan).

FYI
Rao

Readings for Tomorrow..

2008-03-05T15:41:00.001-08:00

Tomorrows agenda is

--> discuss any questions you might have on TCSPs (especially handling of STPs)

--> Discuss next topic: Scheduling

Reading:

Required: (required and in order)

Dana Nau text book 15.1--15.3

If possible, also read
http://rakaposhi.eas.asu.edu/cse574/smith-cheng-slack-aaai93.pdf (6
page paper on a scheduler)

============

Also, after the break, we will have a class to discuss the paper
http://rakaposhi.eas.asu.edu/cse574/KER00.pdf

(which talks about integrating planning and scheduling)

URL for unedited feedback comments...

2008-03-04T17:47:00.000-08:00

http://rakaposhi.eas.asu.edu/cse574/s08-feedback.pdf

They should give you a more complete idea of what the class thinks and whether
it jives with your own view.

If you think the comments have a different meaning that I spun them for,
feel free to tell me that ( http://rakaposhi.eas.asu.edu/cgi-bin/mail?rao

)

regards
Rao

ZENO and Nondeterministic Choices

2008-03-04T16:34:00.000-08:00

In the description of the ZENO algorithm, there is an interesting point (that didn't come up for discussion in class today), which we have covered earlier in the semester. While describing the nondeterministic decisions in the algorithm, the authors list 3 points: decomposing complex goals, choosing actions, and introducing constraints to prevent interference. They then go on to state that completeness requires backtracking on these decisions. That is, just finding a satisficing solution can only be guaranteed if the "nondeterministic" choices are determinised.

However (and this they point out subsequently), we have already seen earlier that for plan-space planners, the order in which the subgoals are selected for satisfaction has no effect on completeness. That is, unless we want to find the optimal solution, we need not determinise (introduce backtrack into) the order in which goals are selected.

This raises some interesting points:

1. Could this dichotomy be seen as a manifestation of the fact that while it is true that *all* subgoals need to be satisfied (at least in non-PSP scenarios) to have *a* solution, a given solution need not contain *all* the possible decompositions, or actions, or constraints?

2. The second part of 1. can perhaps be stated better as follows: any given combination of actions, simple goals and constraints is not necessarily a subset of any given solution, for all solutions in the spaces; however, any given subset of the subgoals is *necessarily* a subset of the subgoal set needed for top-level goal achievement.

These are just some of the things I could think of; I thought this was a very interesting point put forward by the authors. Please feel free to add your own.

kartik

One thought about completeness of ZENO and Decision Epoches planners

2008-03-04T13:49:00.000-08:00

When reading these papers, I'm curious about why ZENO is complete but DEPs are not. One of the reasons I found is the difference between how they model the interaction/ordering between actions in the plans. ZENO post constraints between time points of actions, whereas DEPs uses advancing techniques. More interesting is that, in term of refinement planning, both of them can be considered as refinement operators, and advancing to decision epoches is not a complete refinement.
This difference does not exist in classical planning (i.e comparing POP v.s Progression/Regression) because decision epoches are not there and fattening and advancing work together...
A.Tuan

Final chance: Feedback opportunity to tell me how the class is going

2008-03-03T16:18:00.000-08:00

Folks

Today is your last chance to send the midterm feedback. I will
summarize what I got tomorrow (and discuss in the
class if needed).

I got feedback from 5 students. The class has 12 registered
students--so this is not even a majority.
If it helps you galvanize to action, let me teasingly say that the
majority feedback received till now says
everything is upsy-daisy and requests if any thing for more homeworks and exams.

Rao

---------- Forwarded message ----------
From: Subbarao Kambhampati <rao@asu.edu>
Date: Tue, Feb 26, 2008 at 9:20 PM
Subject: Feedback opportunity to tell me how the class is going
To: Rao Kambhampati <rao@asu.edu>

Folks

Now that 6 weeks are over, I thought it would be good to poll y'all
on what is working and what is not working.
Please feel free to send your comments.

You can send comments either via anonymous web mail
http://rakaposhi.eas.asu.edu/cgi-bin/mail?rao

or by bringing a printed sheet to the class on Wednesday. Note that
the webmail records your ip address--so
you may want to send it from some ip address that is generic.

Here are some things I am interested in finding out. Other comments
welcome too:

1. Are the lectures too fast/too slow/too high level/too low level etc?

2. Are you actually able to connect your readings to lectures?

3. Are you reading before or after the topics are discussed?

4. Should there be more assessment (homeworks/projects etc)?

5. How is the progress towards semester project coming (qualitatively
speaking--
you will have an opportunity for detailed answers anyway)

6. Are the classes interacive enough or should they encourage more discussion?
If the latter, what can we do other than stopping for questions?

7. Overall, how does it compare to other graduate level courses you took/taking
(positively or negatively)

regards
Rao

Agenda an d readings for tomorrow...

2008-03-03T15:56:00.000-08:00

Tomorrow I plan to answer any questions over SAPA and Zeno planning
algorithms (I am hoping you spent time reading them again after last
class), and then shift to temporal constraint networks (which
I mentioned as part of Zeno).

For the first part, I\in addition to the SAPA paper and IJCAI 2007
paper readings that you are already in charge of for last class, you
can also look at the Zeno paper
http://rakaposhi.eas.asu.edu/cse574/zeno.pdf

For the second part (temporal networks) you can look at

http://rakaposhi.eas.asu.edu/cse574/tcn-meiri-dechter-aij.pdf

rao

(New Participation Requirement): Blog discusion after class

2008-02-29T06:01:00.001-08:00

Folks

Starting this week, you are required to make observations on the blog
about things
learned in the class (and readings) every week. These can be questions
you had while thinking
the material over, or skeptical comments or connections between the
week's topics and other areas.

Since we are no longer having note-taking, and have also not been
doing summaries of
papers before the class, I think there hasn't been much discussion of
things done.

In my 4-level classes, I start the discussion with "blog questions"--I
thought this will happen
automatically from your side for this class. And yet, there never seem
to be any questions.

So, I thought some creative pressure should be brought to bear.

Let us see blog posts..

regards
Rao

Semester Project Status/Sumary due next Thursday 3/6

2008-02-28T06:43:00.000-08:00

Folks

A 2-5 page summary and status of your semester project is due in
class on next Thursday 3/6.
I realize that different students are at different stages of progress
in this. Nevertheless, the
3/6 deadline is meant to give you as well as me a clear indication of
where things stand.

In case where you still not zeroed in on a specific topic, I want
your report to give enough details
on what you have done to try to zero in.

thanks
rao

Readings for next class

2008-02-27T11:00:00.000-08:00

Here are the readings for next class

http://rakaposhi.eas.asu.edu/real-temporal-ijcai.pdf

is the main reading. The first part of it re-states the required
concurrency property we talked about yesterday
The second part discusses the decision epoch planning at a high level.

For a more complete description of one decision epoch planner, see
Sections 1 and 2 of
http://rakaposhi.eas.asu.edu/do03a.pdf

Rao

Feedback opportunity to tell me how the class is going

2008-02-26T20:20:00.001-08:00

Folks

Now that 6 weeks are over, I thought it would be good to poll y'all
on what is working and what is not working.
Please feel free to send your comments.

You can send comments either via anonymous web mail
http://rakaposhi.eas.asu.edu/cgi-bin/mail?rao
or by bringing a printed sheet to the class on Wednesday. Note that
the webmail records your ip address--so
you may want to send it from some ip address that is generic.

Here are some things I am interested in finding out. Other comments welcome too:

1. Are the lectures too fast/too slow/too high level/too low level etc?

2. Are you actually able to connect your readings to lectures?

3. Are you reading before or after the topics are discussed?

4. Should there be more assessment (homeworks/projects etc)?

5. How is the progress towards semester project coming (qualitatively speaking--
you will have an opportunity for detailed answers anyway)

6. Are the classes interacive enough or should they encourage more discussion?
If the latter, what can we do other than stopping for questions?

7. Overall, how does it compare to other graduate level courses you took/taking
(positively or negatively)

regards
Rao

Reading for next class (Metric Temporal Planning)

2008-02-22T07:42:00.001-08:00

Read the paper at
http://rakaposhi.eas.asu.edu/cse574/JAIRpddl.pdf
(sections 1-5).

rao

Re: CSE 574: HW 1 Qn IV

2008-02-20T07:47:00.000-08:00

My apologies for the mess-up in the references in Qn IV. I corrected
them now online. Here it is fyi
---------------
Qn IV

For this problem, you will use the planning graph in the last level
III.b above.

IV.a. Convert the planning graph into a CSP encoding (the problem is
small enough that you can write the entire encoding down). Show a solution
for this CSP encoding, and show how it corresponds to a plan.

IV.b. Do IV.a. but with SAT encoding of the planing graph.

IV.c.
Do an "explanatory axiom" (or backward proof based) encoding
of this problem (for the same length as the planning graph you used
in the previous parts). Mark which constraints are similar, different,
stronger etc. compared to IV.b.

-------------------

On 2/19/08, Jonathan.D.Gibbs@asu.edu <Jonathan.D.Gibbs@asu.edu> wrote:
> Okay, so here is my question again.
> In parts a and b of Question 4, we are asked to encode the planning graph from "II.b" in CSP and SAT; part c, meanwhile, asks us create a third encoding and compare its constraints to "III.b." Are these intended to reference the planning graph with mutex-propagation in Question III.b (which is derived for a domain having only 2 actions and 4 variables), or the "standard mutex graph... used by planning graph" in Question II.b (which is derived for a domain with 5 actions and 5 variables)?
>
> -Jonathan Gibbs
> jondg@cox.net
> OR
> Jonathan.D.Gibbs@asu.edu
>

Reading for Replanning--online planning--execution monitoring (next class)

2008-02-20T05:28:00.001-08:00

Here are the readings for next class:

Required

Russell & Norvig: 12.5 & 12.6

Pell et. al. Robust planning and replanning for spacecraft..
http://rakaposhi.eas.asu.edu/cse574/pell97robust.pdf

suggested
Replanning: A new perspective
by Cushing & Kambhampati

http://rakaposhi.eas.asu.edu/replan-will.pdf

Monitoring optimality during execution
by Fritz & McIlraith

http://www.cs.toronto.edu/~fritz/publications/papers/fri-mci-icaps07.pdf

Re: Optional readings for references from today's class

2008-02-15T13:14:00.001-08:00

Should work now.

On 2/15/08, Nan Li <Nan.Li.3@asu.edu> wrote:
> Dear Dr. Rao,
>
> I cannot open the file downloaded from the first link. Could you please
> check that? Thank you.
>
> Best,
> Nan
>
>
>
> On Thu, Feb 14, 2008 at 7:32 PM, Subbarao Kambhampati <rao@asu.edu> wrote:
>
> > Here are optional readings for references made in the class
> >
> > --> A short (4 page) paper on FF (explains the enforced hill climbing
> > and other little changes it makes on top of relaxed plan heuristic)
> >
> > http://rakaposhi.eas.asu.edu/cse574/ff-aimag.pdf
> >
> >
> > -->A (10 page) paper on Fast-Downward (that detects multi-valued
> > fluents masquerading as boolean ones, and uses relaxed planning graph
> > ideas after that detection)
> >
> >
> ftp://ftp.informatik.uni-freiburg.de/documents/papers/ki/helmert-icaps04.pdf
> >
> > --> A short (3 page) paper on Macro-FF, that selectively learns
> > macros on top of FF
> >
> http://www.cs.ualberta.ca/%7Eadib/Research/Planning/ipc4Alberta.pdf
> >
> > --> A paper on learning search control rules from failures using EBL
> >
> > http://rakaposhi.eas.asu.edu/ebl-po-aaai94.pdf
> >
> > (a longer but perhaps cleaner description is
> >

http://rakaposhi.eas.asu.edu/ebljpap.pdf
> >
> > also, for relations between EBL, dependency directed backtracking in
> > CSP (no-good learning) as well as planning, see
> >
> > http://rakaposhi.eas.asu.edu/jour-ddb.pdf )
> >
> >
> > Rao
> >
>
>

Reading for next class (over-subscription/Partial Satisfaction planning)

2008-02-14T18:43:00.000-08:00

For next class, please read

http://rakaposhi.eas.asu.edu/psp-aaai04.pdf

Rao

Optional readings for references from today's class

2008-02-14T18:32:00.001-08:00

Here are optional readings for references made in the class

--> A short (4 page) paper on FF (explains the enforced hill climbing
and other little changes it makes on top of relaxed plan heuristic)

http://rakaposhi.eas.asu.edu/cse574/ff-aimag.pdf

-->A (10 page) paper on Fast-Downward (that detects multi-valued
fluents masquerading as boolean ones, and uses relaxed planning graph
ideas after that detection)

ftp://ftp.informatik.uni-freiburg.de/documents/papers/ki/helmert-icaps04.pdf

--> A short (3 page) paper on Macro-FF, that selectively learns
macros on top of FF
http://www.cs.ualberta.ca/%7Eadib/Research/Planning/ipc4Alberta.pdf

--> A paper on learning search control rules from failures using EBL

http://rakaposhi.eas.asu.edu/ebl-po-aaai94.pdf

(a longer but perhaps cleaner description is

http://rakaposhi.eas.asu.edu/ebljpap.pdf

also, for relations between EBL, dependency directed backtracking in
CSP (no-good learning) as well as planning, see

http://rakaposhi.eas.asu.edu/jour-ddb.pdf )

Rao

Instructions for submission of mini-project 1

2008-02-13T15:24:00.000-08:00

Please bring a hard-copy report on the mini-project 1 to the class.

In addition, you will also be asked to upload the pddl files of your
domains onto the class wiki.

rao

Tomorrow we will do a lecture on learning techniques for planning--Reading enclosed..

2008-02-13T15:10:00.001-08:00

Folks

In a change of plan, I decided that it would be good to do some
discussion on the status of learning techniques--specifically those
aimed at
learning additional domain knowledge to improve planning performance.
This is something that was in the back of our minds as we kept talking
about HTNs, TLPlan rules
etc.

Although the topic of learning techniques for planning, will, I hope,
make a comeback again during the semester, this is also a good time to
do spend one lecture on it.

Accordingly, the readings for tomorrow's class are:

Sungwook Yoon's paper on learning to improve relaxed plan heuristics:

http://www.public.asu.edu/~syoon10/icaps06.pdf

You can also look at the tutorial on Learning Techniques for Planning:
http://rakaposhi.eas.asu.edu/learn-plan.html

(there is even a video of an earlier version of this tutorial at
http://videolectures.net/mlss06au_kambhampati_ltp/ )

Rao

Re: Readings for the next class

2008-02-12T05:07:00.001-08:00

The readings are still Chap 11 and Chap 10.

We continue discussion of HTN planning (chap 11) and then do control
rules--especially
temporal control rules (chap 10)

rao

On 2/11/08, Aishwarya Sivaraman <asivaram@asu.edu> wrote:
> Dr. Rao,
>
> I was just wondering if you have put up the readings for tomorrow's class. I
> could not see any updates on the blog.
>
> Best Regards,
> Aishwarya Sivaraman
>

Homework 1 released. Due in 2 weeks.

2008-02-10T08:50:00.001-08:00

Homework 1 is available from the home page (homeworks tab)

It covers everything done until last Tuesday

It will be due back in 2 weeks.

Please take homeworks seriously; there is a possibility that we will decide
not to have separate exams.

Also, please note that you are required to do the
work yourself and not use any old solutions that you may have been
able to download
from the web. It is okay to discuss with other students, instructor
etc, but the final work must
be done by you.

Rao

Fwd: Broken link of paper mentioned in last class

2008-02-07T06:06:00.001-08:00

sorry. the link should work now..

http://rakaposhi.eas.asu.edu/AMali99.pdf]

The slides for the talk accompanying the paper are at

http://rakaposhi.eas.asu.edu/mali-aaai99/index.htm

Rao

On Feb 7, 2008 6:17 AM, Tuan A. Nguyen <Tuan.Anh.Nguyen@asu.edu> wrote:
> Dear Rao,
> Your paper "On the utility of plan-space (causal) encoding" can not be
> downloaded from your homepage.
> Best regards,
> A.Tuan
>

Readings for Next class: Chapters 11 and 10 in Ghallab/Nau/Traverso textbook

2008-02-06T09:31:00.000-08:00

Folks

There will be a slight change of plan for next class. Instead of
doing partial satisfaction planning, we will do
"knowledge-based" planning--with some emphasis on hierarchical task
network planning. The readings for the
class are Chapter 11 and then Chapter 10 (in that order) from Ghallab/Nau book.

Rao

the encodings paper --working URL

2008-02-03T19:14:00.001-08:00

Thad pointed out that the URL for the encodings paper is broken.

Here is the working URL
http://citeseer.ist.psu.edu/64263.html

(the reason the oldone is broken is that citeseer at nec is no longer working)

rao

Readings for next class (as well as additional readings for yesterday)

2008-02-01T05:28:00.000-08:00

Assuming you have already read the readings for yesterday's class,
here are additional readings for next class

http://rakaposhi.eas.asu.edu/cse574/readings.html#encodings

The second and third papers--CSP encodings and SAT encodings--are most
relevant for next class.

The first one explains the details of how to use EBL in graphplan
(that I described at a highlevel yesterday).

rao

Response to first thinking topic

2008-01-31T19:13:00.001-08:00

If the state space is not a tree, there is a cycle, and so running around in the cycle as much as desired before achieving the goal generates plans of any desired length.

That being said, there tends to be very little interesting in generating plans with cycles, so one can always prune any plan with 2^k actions (which visits 2^k+1 states -- 1 more than the total number of states, ergo, the plan is cyclic, removing the cycle shows that it is non-minimal).

More practically, one keeps a closed list, i.e., a hash table of visited states. There is never a reason to pass through a real state or a regressed "state" in more than the shortest distance, so one tracks that number and prunes any search branches failing this condition. In some formalisms this is expensive to do, so instead one simply examines the states visited by the current branch, and prune the branch if there are any cycles.

Removing such restrictions does allow progression and regression planners to find any walk through state space, i.e., all plans minimal or otherwise. Such planners could enumerate all such walks, if desired, by printing plans whenever the goal check succeeds (but refusing to terminate, and generating the children of the state even though the goal is already true), but the search control does have to be breadth or best first [i.e., there must exist some measure of a plans with the guarantee that solutions are printed in sorted order with respect to the measure, and that only finitely many plans ever map to a value of the measure]. These considerations are important to partial satisfaction planning, actually.

Regression can, of course, generate cycles as easily as progression. One can almost invert any STRIPS domain; the inverted domain requires disjunctive and negative preconditions (and the original domain has to insist on conjunctive and positive preconditions). The idea here is that progression in the inverted domain is exactly regression in the original domain [so general properties of progression are also general properties of regression, and vice versa, unless the properties depend upon purely conjunctive or purely positive preconditions].

Minimizing a plan, by this definition, is going to be exponential -- every subset of actions would have to be checked in worst-case scenarios. In many domains, however, a simple search considering removing one action at a time could terminate early in every branch, leaving polynomial time. (if, for example, each action contributes the truth of a proposition that is needed by another action known to be necessary, and that no other action in the plan gives that condition, then this action is also necessary (counting the initial state and goal states as actions, with the goal state being the only action initially marked necessary). A simple backwards sweep is sufficient. Slight optimizations would be sufficient to minimize Blocksworld plans, and i think the approach as stated works for Rovers plans without cycles in move operators [which are easy to separately detect])

Different notions of planning define minimality in different ways; progression and regression normally imagine minimality to mean acyclic. Under that notion of minimality, with normal sorts of pruning based on cycle checking or duplicate detection, progression and regression generate minimal plans.

Under the suggested definition of minimality, progression and regression can generate non-minimal plans under any search control that doesn't guarantee optimality [even if employing duplicate detection or cycle elimination]. First note than an optimal plan must be minimal, since removing any set of actions reduces cost. Consider the following sketch of a state space:

sI -A-> s1 -B-> s2 -A'-> s3 -Z-> sG
-A'-> s3 -Z-> sG

An optimal planner will visit s3 by the shortest path first (optimal planners typically guarantee optimality in this recursive manner). A suboptimal planner, however, might choose to do ABA'Z, which is acyclic and non-minimal (A and B can be removed). Note that if B depends upon A, and A' depends upon B (or the initial state, and A deletes the condition that both the initial state and B assert), then removing either A or B alone leaves an unsound plan [thus showing the need to consider all subsets of a plan].

A planner that only prunes cyclic search branches (instead of keeping a closed list) will generate such acyclic, non-minimal, sub-optimal solutions with the greatest of ease.

For a much more thorough analysis of a very similar problem, one could look at Backstrom's analysis of reordering and deordering of plans; reordering being the hard problem of determining whether or not a plan has many alternative causal-link proofs [this is related to minimality as defined here].

While optimality implies minimality, and non-minimality implies sub-optimality, minimality does not imply optimality. (Under either definition of minimality: acyclic or every subset kills the plan).

-Will

A screed and discussion on the optimality track of planning competition...

2008-01-31T16:20:00.001-08:00

You might want to check out

http://raos-ruminations.blogspot.com/2006/07/on-suboptimality-of-optimal-planning.html

for a recent (as in 1.5 years back controversy) about optimality track
of the IPC

Rao

Email addresses of initial advisors

2008-01-29T19:54:00.001-08:00

In case you don't know how to get in touch with initial advisors

wcushing@asu.edu
j.benton@asu.edu
sungwook.yoon@asu.edu
menkes@asu.edu
rao@asu.edu

rao

Readings for next class + additional material for todays'

2008-01-29T16:31:00.000-08:00

Folks

As I said, we will discuss any outstanding questions on heuristics
(including probably a quick discussion on use of reachability
heuristics for
Plan-space planning), and then discuss bounded-length planning as a
combinatorial problem. What we did with Graphplan today will become
a simple subcase--with other cases being pushing planning as SAT/CSP/IP etc.

The additional reading for tomorrow's class is:

Chapters 6 and 7 in Nau et. al.'s text book

or if you want a shorter version, Chapter 11 (sections 4 and 5) in
Russell and Norvig (if you don't have the text, here is that chapter
from AIMA site:
http://aima.cs.berkeley.edu/newchap11.pdf )

-----------------

Also, if any of you felt a little lost today and wanted a more
structured lecture on the heuristics, check out

http://rakaposhi.eas.asu.edu/ICAPS06-tutorial/ICAPS06-1.mp3

which is the first part of the tutorial based closely on the paper you
read and slides you saw.

Rao

Initial advisors for cse574 projects... You will have to meet with your advisor by end of this week..

2008-01-29T11:09:00.001-08:00

Folks

Thanks for writing down your initial ideas for class projects. We
(myself and the Teaching czars) went through them and
put in our bids.

Please look at https://wiki.asu.edu/planning/index.php/Project_ideas

and see who your primary advisor bid is from.

You will have to contact the advisor and make an appointment (about an
hour or less in most cases) so you can talk to them
and flesh out what is potentially feasible.

You will be required to maintain a wiki page with the summary of the meeting.

Rao

Reading for next class--reachability heuristics for planning

2008-01-24T17:55:00.001-08:00

For the next class, read the following paper

http://rakaposhi.eas.asu.edu/pgSurvey.pdf

(definitely upto page 61,
and preferably upto page 66. Before you die of shock,
note that the paper starts at page 47 ;-)

No summaries are needed--but you should read the paper before coming.
Since part of this material has been covered once in
471 (if you took it from me), I intend to go at a brisk pace (brisk
pace for rao is not unlike realtime for redwoods).

Rao

Fwd: A little more on white-knight clause..

2008-01-22T21:44:00.001-08:00

After class, Tuan and Kartik had a ton of questions on the infamous
white-knight clause. Here is a gist of that discussion for general
edification. (Tuan--you can copy this over to your scribe notes as
necessary).

The term white-knight comes from the "White knights in shining armor
rescuing damsels in distress".. The white-knight step intervenes
between the threatening step and the step needing the condition and
re-establishes the condition.

*NOTE* that while white-knight clause is needed to prove correctness
of arbitrary partial order plans, it is *NOT* needed to ensure
completeness of plan-space planning (which, like progression and
regression, only needs to generate all action sequences that can solve
the problem). While the plan-space planner I discussed in the class
will never be able to terminate with the white-knight example plan
(because there is no way to come up with an un-threatened causal link
for that plan), it is able to terminate with two different
specializations (refinements) of that plan--such that every action
sequence that is a top-sort of the white-knight plan is also a
top-sort of one of these refinements.

*IF* for some strange reason, you do want your plan-space (partial
order) planner to terminate with the white-knight plan, then you need
to generalize the semantics of causal links such that you have three
ways of resolving a threat. Specifically, if a causal link s1--p--s2
is threatened by a step s3 (which has effect ~p), then, we have three
resolution possibilities:

1. Promotion: put s3<s1
2. Demotion: put s2<s3
3. White-knight "de-clobbering"
Find a step s4 (either existing in the plan, or from the library)
such that s4 has effect p.
Introduce ordering s3<s4<s2
(so s4 comes between s3 and s2)

As mentioned above, the white-knight declobbering is unnecessary for
completeness in the space of action sequences.

It is also worth noting that when you allow white-knight declobbering,
you are essentially changing the semantics of the causal-link
constraint. Without declobbering, we in essence say that once
established, the causal link condition *CANNOT BE VIOLATED EVEN
TEMPORARILY*. Allowing declobbering means you can allow it to be
temporarily violated.

[Maintenance goals] Sometimes, plan-space planners use causal links to
model *Maintenance* goals. (The normal goals we saw until now are
goals of attainment--i.e., the conditions must be true at the end.
Sometimes, you also have goals of maintenance--all through the
execution of the plan, a certain condition must hold true. For
example, if you want to walk through a dark basement garage to the
fuse box to reset the fuse, then you need light to help you navigate
the garage. This light must be on pretty much all through the
execution of the plan. This kind of semantics will not be respected by
white-knight declobbering.

[White-knight and Chapmans Modal Truth Criterion;] White-knights
entered planning vocabulary through an (unfortunately) influential
paper from 1987 by David Chapman.

I stored a copy at http://rakaposhi.eas.asu.edu/cse574/notes/chapman-mtc.pdf

for your edification

This paper was unfortunate in that it (1) confused plan-space planning
with the need for being complete in the space of partial plans (which
as I pointed out is *NOT needed*) and (2) consequently assumed that
per-node complexity in plan-space planning will be influenced by the
cost of checking correctness of a partially ordered plan. Another
wrong assumption that Chapman made was that plan-space planners *MUST*
deal with partially instantiated partial order plans (this is the
*lifted planning* I started talking about).

Since a PO Plan can have an exponential topoligical sorts, and many
more ground topological sorts, it is not surprising that checking the
correctness of all those topological sorts can potentially be
exponential.

Chapman's +ve result was that *if* actions have no conditional
effects, and all variables are considered to have *infinite* domains
(which is a bogus assumption), then correctness can be checked in poly
time (note that this requires the white-knight clause). His negative
result was that once you have conditional effects, the correctness
check will be NP-complete. He somehow made this sound as if it was
the end of the world (and the planning community, for its part, went
along with it). On scholar.google.com, you see that Chapman's paper
has had 845 citations--most before 1991..

http://scholar.google.com/scholar?q=chapman+modal+truth+criterion&hl=en&lr=

The third paper in that page--written some 7 years after Chapman's
paper--brings up some of these critiques.

Rao

ps: What is wierd about Chapman's paper is that Chapman should have
known better. 10 years before his work, the planner "Nonlin" was
developed by Austin Tate and it saw farther (the plan-space planner we
discussed today is closer to 1977 paper by Tate than anything in
Chapman's 1987 paper). Apparently Chapman did talk to Tate but somehow
Tate's work is not cited prominently. Checkout

http://rakaposhi.eas.asu.edu/nonlin-review.txt

for an opinionated review of this whole episode..

Refinement Planning Mandatory reading for next class (summary required before class--post as a comment on the blog)

2008-01-22T18:29:00.000-08:00

Here is the paper you have to read for next class. You will be
required to write and post a summary of the paper as a comment to this
posting on the blog

http://rakaposhi.eas.asu.edu/kambhampati.pdf

Rao

Policies on scribing..

2008-01-22T09:44:00.000-08:00

Folks:
Unless otherwise stated, the deadlines for the scibe notes to be put
on the wiki are:

Tuesday class: Ideally by Thursday---but definitely by Saturday

Thursday class: Ideally by Saturday--but definitely by Sunday

This will give us czars a chance to make changes to the notes if needed.

If you see yourself falling behind this schedule, it is your
responsibility to let us know.

regards
rao

[Thinking topic]: Completeness and minimality..

2008-01-21T14:31:00.001-08:00

[Part of the required participation in the class involves taking part
in the blog discussions--and giving answers to thinking topics. Here
is one]

An interesting issue about planner completeness is whether or not the
planner is capable of finding *every plan* for the problem.

Let us get some terminology straight.

A sequence of actions a1...an is considered a solution to a problem
[I,G] if we can
execute the sequence starting in state I and G will hold in the final
state after executing an.

A solution plan P: a1...an is said to be minimal if it is not
possible to remove any subset of actions from P
without violating its solution-ness.

An example of a non-minimal plan for achieving On(A,B) when A, B and C
are all on table is:

put A on C, take A of C, put A on B.

Cleraly, we can cut the first two actions out and it will still be a
solution (so it is a non-minimal solution).

Consider the following questions:

0. Given a planning problem that is solvable, how many solutions
(minimal as well as non-minimal) are there?

1. Is progression planner complete for all plans (including minimal
and non-minimal plans)?

2. Is regression planner complete for all plans (including non-minimal)?

3. Will regression planner ever find any non-minimal plans?

4. Will progression planner ever find any non-minimal plans?

5. If a planner is capable of producing non-minimal plans, then will
sussman anomaly be non-serializable for such a planner?

6. Given a possibly non-minimal plan, how costly would it be to
"minimize" it (i.e., remove redundant actions from it)?
(Complexity-wise... is it polynomial?
exponential? etc)

Mailing list and class blog set up..

2008-01-21T13:43:00.001-08:00

Folks:

If you are receiving this mail, you are still registered for CSE574
(i.e., forgot to drop the class before add/drop end ;-).

I created a mailing list as well as a blog for the class. Most of you
know the drill:

The intended usage of the mailing list is for me to send you
announcements as well as things to think about.
All mails sent to the mailing list are sent to your email address,
posted to the class mail archive as well as to the class blog.

All of you have posting privileges to the class blog. You can post to
the blog as well as write comments on other posts.
You can keep track of changes to the blog through RSS feed etc.

cheers
Rao

CSE574 Planning & Learning

Readings for link-analysis lectures...

Class evaluations for your (temporary) edification..

The Elusive cse574 Cumulative Grade Book

Here is the link for all the semester project reports...

an old midterm and some homework solutions..

your peer reviews.. all tf and no idf...

Please reply to this mail with your paper and (optionally) presentation as attachments

bring your presentations on a memory stick..

Fwd: Real world blocksworld

Correction for a homework 2 problem..

Assistance or distraction?

*Important*: Format of the class of 4/29--you will do short presentations of your project findings

(correct URL) Reading for tomorrow (as well as slides and audi...

Reading for tomorrow (as well as slides and audio from today's talk)

Fwd: teaching evaluations

Refreshed the plan recognition slides from the first plan recognition lecture..

Reminder: Talk Today 11AM on "The Myth and Reality of Web Services Composition" by Biplav Srivastava of IBM research 4/23 11AM BY 510

Aiming to minimize the expected fallout from the unexpected hanging... ;-)

Talk on "The Myth and Reality of Web Services Composition" by Biplav Srivastava of IBM research 4/23 11AM BY 510

talk on Online Continual Planning to Control Modular Production Printer (Minh Do; PARC Labs) 4/24 3:15pm BY 190

Re: [CSE574 Planning & Learning] readings for next class

Re: [CSE574 Planning & Learning] readings for next class

readings for next class

Link to the goal graph based plan recognition discussed in today's class..

an efficient temporal planner for handling required concurrency..

Reference on McLug

Reading for next class

(IMPORTANT) Schedule for the rest of the semester etc.

Readings for Tuesday's class

Homework 2

Cooperative planning using multiple agents

Plan (and readings) for next Tuesday's class (4/8)

Readings for Tomorrow--LAO* and LRTDP

Reminder: Reading for next class: Decision Theoretic Planning

Fwd: [CSE574 Planning & Learning] Reminder: Seminar today by Dana Nau (11:30AM BY2...

Reminder: Seminar today by Dana Nau (11:30AM BY210) on "Planning for Interactions among Autonomous Agents"

Need for assessment in the class??--feedback requested (can be on the blog)

Over-sensing and LCW actions.. (as well as planning with sensing as internet planning)

Reading for next class: Decision Theoretic Planning

Re: A short little paper for next class reading..

A short little paper for next class reading..

Added several slides to yesterday's lecture

The references for the complexity results on temporal planning and planning with partial observability/non-determinism

Why is this still a problem?

Reading for Thursday's class (belief states...)

Is CSP scheduling better than IP scheduling?

Planning and Scheduling

Re: CSE574 grades

Slack-Based Heuristics Scheduling Paper

Added slides on quotienting and candidate set semantics for temporal planning

Readings for Tomorrow..

URL for unedited feedback comments...

ZENO and Nondeterministic Choices

One thought about completeness of ZENO and Decision Epoches planners

Final chance: Feedback opportunity to tell me how the class is going

Agenda an d readings for tomorrow...

(New Participation Requirement): Blog discusion after class

Semester Project Status/Sumary due next Thursday 3/6

Readings for next class

Feedback opportunity to tell me how the class is going

Reading for next class (Metric Temporal Planning)

Re: CSE 574: HW 1 Qn IV

Reading for Replanning--online planning--execution monitoring (next class)

Re: Optional readings for references from today's class

Reading for next class (over-subscription/Partial Satisfaction planning)

Optional readings for references from today's class

Instructions for submission of mini-project 1

Tomorrow we will do a lecture on learning techniques for planning--Reading enclosed..

Re: Readings for the next class

Homework 1 released. Due in 2 weeks.

Fwd: Broken link of paper mentioned in last class

Readings for Next class: Chapters 11 and 10 in Ghallab/Nau/Traverso textbook

the encodings paper --working URL

Readings for next class (as well as additional readings for yesterday)

Response to first thinking topic

A screed and discussion on the optimality track of planning competition...

Email addresses of initial advisors

Readings for next class + additional material for todays'

Initial advisors for cse574 projects... You will have to meet with your advisor by end of this week..

Important: Format of the class of 4/29--you will do short presentations of your project findings