ITEE Ph.D confirmation seminar: Penny Sweetser, 11.00AM, Mon 28 Apr 2003

Intelligent Command Agents for Command and Control Simulations using Cognitive Work Analysis and Neural Networks

Speaker: Penny Sweetser, ITEE

When: 11.00AM, Monday 28 Apr 2003

Venue: 78-420

Host: A/Prof Janet Wiles

Abstract:

  Simulations are used in the military for training, planning,
  rehearsal and analysis. Intelligent agents are needed in these
  simulations to fill the roles of military personnel, who can be
  costly or unavailable. Current agents in these simulations are
  designed using methods of task analysis or naturalistic decision
  making. However, these methods give rise to agents that are
  predictable, rigid, unprepared for unforseen situations and that are
  far from human-like. Previous work at UQ, in conjunction with the
  DSTO, has used cognitive work analysis to define the domain and
  actions of agents in military simulations. This work has created
  agents that can act intelligently in a much wider variety of
  situations, not just the set of normal, expected situations. The
  command agent in this system uses rules to support a
  Belief-Desire-Intention (BDI) model, derived using cognitive work
  analysis and military doctrine, to choose a course of action to
  take. However, it is not possible to derive rules for the weighting
  of different factors that need to be considered for each decision,
  such as surprise, flexibility and administration. When human
  commanders choose a course of action they weight these factors based
  on the current situation and previous experience. There is no
  doctrine or rules that cover which factors are more important and
  human experts weight these factors without following a specific
  process. This project will investigate the use of different machine
  learning techniques, including linear regression, decision trees and
  neural networks, to weight these factors and compare their ability
  to be used in conjunction with, and improve, the current
  model. These techniques will be trained on data from human players,
  assessed for ability to learn and predict this data, and implemented
  and tested in the simulation.

Biography:

(biography unavailable)

Type:

Ph.D confirmation

Contact:

A/Prof Janet Wiles, seminar host (janetw@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator) (guido@itee.uq.edu.au)

ITEE seminar web page: http://www.itee.uq.edu.au/~seminar

[All seminars]