Hector Geffner's: "Reminiscences of Influential Papers in Planning"

[Subbarao Kambhampati <rao@cotopaxi.eas.asu.edu>]

[[**don't reply to this address**]]

[Rao: Back in July, I asked for people to share their views on papers they were
influenced by in planning. 
(see http://rakaposhi.eas.asu.edu/planning-list-mailarchive/msg00042.html )

As you know, we are much better at planning  than execution :-) 
Here then is the first contribution. 
Now that the ice is broken perhaps others will join..]


==========mail from hector.geffner@tecn.upf.es--unedited as promised====

>From: hector.geffner@tecn.upf.es
>Date: Thu, 21 Nov 2002 14:36:33 -0500
>Reply-to: hector.geffner@tecn.upf.es

Hi Rao,

I just came across your call for reminiscences through a google search(!);
apparently you dropped me off the list (no, actually, the msgs probably
are being sent to an old address of mine where they are filtered; can you
use my new address? thanks!)


I dont know whether im old enough to have reminiscences but since i got
into planning not long ago, i still remember the papers that were influencial
for me, and still are, in particular:

Andrew Barto, Steven Bradtke and Satinder Singh. Learning to act using
real-time dynamic programming. Artificial Intelligence 72 (1995),
pp. 81-138.

In the page of "positive reviews" about Logical-AI coordinated by Vladimir
Lifschitz, I wrote back in 1997:  ".. The paper by Barto et al is not
about specifying dynamic systems but about using dynamic systems for selecting
optimal courses of actions. The paper is beautifully written and is mostly
self-contained, and provides a unifying theoretical foundation for a number of
crucial problems in AI related to action, including planning, learning and
control. It's not logical AI, but still AI at its best... "

Back then we also found Korf's paper on LRTA* quite illuminating, and borrowed
some intuitions also from Pattie Maes "A Spreading Activation Network for
Action Selection". Basically, we replaced the "spreading activation" by an
"heuristic activation" and used it in a greedy/LRTA* search .. these
are the basic ideas of our first paper on planning at AAAI97: "A fast and
robust action selection mechanism for planning" ...

-hector