Accurately determining intermediate and terminal plan states using bayesian goal recognition

David Pattison, Derek Long

Research output: Chapter in Book/Report/Conference proceedingConference contribution book

32 Downloads (Pure)

Abstract

Goal Recognition concerns the problem of determining an agent's final goal, deduced from the plan they are currently executing (and subsequently being observed). The set of possible goals or plans to be considered are commonly stored in a library, which is then used to propose possible candidate goals for the agent's behaviour.
Previously, we presented AUTOGRAPH - a system which removed the need for a goal or plan library, thus making any problem solvable without the need to construct such a structure. In this paper, we discuss IGRAPH, which improves upon its predecessor by utilising Bayesian inference to determine both terminal and intermediate goals/states which the agent being observed is likely to pass through.
Original languageEnglish
Title of host publicationGAPRec 2011. Proceedings of the First Workshop on Goal, Activity and Plan Recognition
Subtitle of host publicationICAPS 2011
EditorsDavid Pattison, Derek Long, Christopher Geib
Pages32-37
Number of pages6
Publication statusPublished - 12 Jun 2011

Keywords

  • goal recognition
  • plan recognition
  • planning
  • intermediate
  • terminal plan states
  • bayesian goal recognition

Cite this

Pattison, D., & Long, D. (2011). Accurately determining intermediate and terminal plan states using bayesian goal recognition. In D. Pattison, D. Long, & C. Geib (Eds.), GAPRec 2011. Proceedings of the First Workshop on Goal, Activity and Plan Recognition : ICAPS 2011 (pp. 32-37)