An Approach to Planning in the Inflight Emergency Domain.
Abstract
A planner must understand its domain and be able to effectively reason about interacting goals competing for satisfaction in its environment. Current artificial intelligence (AI) planning structures are inadequate for expert and common sense reasoning in the dynamic aircraft inflight emergency domain. These current planners are inadequate because they are not designed to manipulate multiple goals in an unpredictable environment. Conflicting goals, i.e., when the realization of one goal interferes with the realization of another goal, poses particularly frustrating problems for typical planners. Like the everyday activities domain, the flight domain requires a large pool of world and common sense information. It also requires flight domain information and knowledge of the goals of its pilots and other aircrew members. The planner for an intelligent pilot aid (PIPA) divides planning into four activities, (1) goal detection, (2) plan proposition, (3) plan projection, and (4) execution; composed into four components with similar names. Goals are associated with the observations which trigger them, and plans are associated with the goals to which they apply. Proposed plans are simulated in a hypothetical model of current states and are watched for weakness, overlap, or conflict. Overlapping plan steps are approximately combined, while conflicts direct the PIPA to either prospose and test plans or have the PIPA attempt to change the circumstances surrounding the conflict. Originator supplied keywords include: Artificial Intelligence, Pilots, Planning, Reasoning, Emergencies, Theses.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1984
- Accession Number
- ADA151869
Entities
People
- D. L. Knode
Organizations
- Air Force Institute of Technology