Automated planning, the ability to synthesize a course of action to achieve desired goals, is an integral part of intelligent agency. While there have been tremendous strides made in the expressiveness and efficiency of planning systems, most work still views planning as pure inference over completely specified domain models and objectives. In many real world scenarios however, the domain models as well as the objectives are only partially specified. For example, in a military planning scenario, domain models are incomplete due to dynamic environments as well as lack of knowledge about the adversaries. Similarly, preference models are incomplete since the commanders are often unable to articulate full tradeoffs guiding their objectives. The traditional planning frameworks that expect complete and correct models up front are unsuitable for the job. Instead, we need to develop a planning framework that is able to get by with incomplete models pending their completion. In this project, we propose to develop a comprehensive planning framework for plan synthesis in the presence of incomplete domain and preference models. This raises several critical challenges (i) what knowledge can be used to circumscribe the incompleteness and (ii) what are the solution concepts for plans in the presence of incomplete domain and preference models? (iii) how to efficiently syntehsize plans that satisfy these solution concepts? We made some preliminary progress twoards these aims, specifically for the case of imprecise objectives. In the current project, we plan specifically focus on robust planning with incomplete domain models. We will consider two types of knowledge sources to complement the incomplete models: (i) incompleteness annotations and (ii) library of previous cases. In both cases, we develop principled techniques for efficiently generating robust plans. We also propose to investigate methods that can simultaneously handle objective and model incompleteness. Finally, we will look at techniques for learning to reduce model and objective incompleteness over long term. Planning technology developed in this research is expected to significantly increase the reach and application of automated planning techniques. It will be especially useful in many military planning domains, where the modeling burden and dynamically changing and/or partially articulated planning objectives make traditional planning technology unsuitable. It will also be useful in web-service composition and workflow management scenarios. 3
|Effective start/end date||1/1/13 → 3/30/17|
- DOD-NAVY: Office of Naval Research (ONR): $450,000.00
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.