Preferences and Partial Satisfaction in Planning

[AAAI 2010 Tutorial]

J. Benton*|

Jorge Baier|

Subbarao Kambhampati*

Planning research has traditionally focused on scenarios in which all stated goals are treated as hard constraints that must be met to satisfy the problem. However, many real-world planning applications often require the consideration of user preferences. Preferences, unlike hard goals, are properties of a plan that are desired in a solution, but not enforced. Example applications range from robotics (in which robots are expected to respect as much as possible their user's preferences) and software applications (e.g., component software composition, web service composition). Preferences are also useful in applications in which certain goals are allowed to remain unachieved due lack of resources. In recent years, much attention has been given to solving these types of problems within the planning community which involve issues of plan quality optimization, as opposed to only finding a satisficing plan. This new interest is evidenced by the recent preference and net benefit tracks in the respective 2006 and 2008 International Planning Competitions. Furthermore, much recent research has begun to focus on solving for temporally extended preferences over a state trajectory, handling preferences in HTNs and in partial satisfaction planning on which the net benefit competition track was based.

In this tutorial, we will examine the current state-of-the-art in preference-based, partial satisfaction planning and will cover representation of preferences and soft constraints, defining objective functions, frameworks for solving for preference-based and over-subscription planning problems, and current open challenges in preference and over-subscription planning

Tutorial Material

Subbarao Kambhampati

You are visitor number to this page!
Last modified: Mon Aug 20 11:37:20 MST 2012