- Abstract:
-
"Plan recognition" is the identification of an agent's plans on the basis of observations of the agent's actions. The vast majority of artificial intelligence research in this area has made the assumption that agents were not actively hostile to others inferring their plans. As a result, there are a number of open research problems in plan recognition for adversarial domains. These problems can be divided into two classes: problem complexity (e.g. partially observability, inference about actions that are not observed ("the alarm didn't sound"), and the inference of multiple concurrent goals) and misdirection (e.g. feints and bluffs).
In this chapter, we will discuss these issues, some artificial intelligence approaches for plan recognition, and their limitations. We will then discuss a specific hybrid logical and probabilistic approach to plan recognition embodied in the Probabilistic Hostile Agent Task Tracker (PHATT). We will discuss how this approach deals with some of the problems presented by adversarial domains, discuss lessons learned, and suggest areas for future research.
- Links To Paper
- 1st Link
- Bibtex format
- @InProceedings{EDI-INF-RR-0963,
- author = {
Christopher Geib
and Robert Goldman
},
- title = {Recognizing Plan/Goal Abandonment},
- book title = {International Joint Conference on Artificial Intelligence},
- publisher = {Morgan Kaufman},
- year = 2003,
- pages = {1515-1517},
- url = {http://dli.iiit.ac.in/ijcai/IJCAI-2003/PDF/258.pdf},
- }
|