Lectures delivered at the ACAI Summer School on Automated Planning and Scheduling, June 2011

Abstract:

In its early days, the planning community routinely and gleefully let its reach exceed its grasp in terms of the class and scope of problems under consideration. Even when our planners were really classical but quite glacial, and could at best handle three blocks problems under mere minutes on a good day, we still blithely directed myriad efforts at lifted planning, temporal planning, stochastic planning, open world planning, mixed-initiative planning, and multi-agent planning.

The principled scale-up in classical planning in the last decade should have opened a more expansive vent for all that pent-up ambition. Alas, it hasn't quite turned out that way; our successes in scale-up seem to have turned us more circumspect. A Martian looking at any of the recent ICAPS proceedings can be forgiven for thinking that we are all mostly in quest of ever-more speed-up for classical planning.

In these lectures, I will make a case for turning our (and especially your) energies back to the future of planning, and explain how we can co-opt the scale-up in classical planning to aid in this quest. We shall look, in particular, towards advances in partial satisfaction planning, temporal planning, stochastic planning, as well as planning with incomplete models and open worlds.

Slides (final version; as delivered)

Audio Part 1

Audio Part 2


Subbarao Kambhampati
Last modified: Fri Jun 17 15:47:21 MST 2011