You are here

Genetic Programming-based Agent Training System (GPATS)

Award Information
Agency: Department of Defense
Branch: Army
Contract: N61339-02-C-0033
Agency Tracking Number: A012-1864
Amount: $70,000.00
Phase: Phase I
Program: SBIR
Solicitation Topic Code: N/A
Solicitation Number: N/A
Timeline
Solicitation Year: N/A
Award Year: 2002
Award Start Date (Proposal Award Date): N/A
Award End Date (Contract End Date): N/A
Small Business Information
725 Concord Avenue
Cambridge, MA 02138
United States
DUNS: 115243701
HUBZone Owned: No
Woman Owned: No
Socially and Economically Disadvantaged: No
Principal Investigator
 Karen Harper
 Principal Scientist
 (617) 491-3474
 kharper@cra.com
Business Contact
 Paul Gonsalves
Title: Vice President
Phone: (617) 491-3474
Email: pgonsalves@cra.com
Research Institution
N/A
Abstract

"Increasingly sophisticated models and simulations developed by DoD to support a range of analysis and training functions have triggered a strong demand for realistic representation of the key human decision makers that populate the battlespace. Developingadaptive and robust human behavior representations (HBRs) has not been a trivial exercise, however, especially given the dependence of most state-of-the-art HBR approaches on intensive domain-specific knowledge acquisition (KA) and manually-mediatedknowledge representation (KR) efforts. To address these problems, we propose to develop a Genetic Programming-based Agent Training System (GPATS) that will employ an evolutionary computation approach that will train agents to combine basic skills in anoptimal fashion, thus generating more complex individual skills and emergent group behaviors that will prove effective in military operations. Our approach includes three primary components: 1) a set of "seed functions" that describe simple behaviors thatour agents possess a priori (e.g., walking, carrying a weapon, communicating); 2) a Genetic Programming (GP) engine that employs an evolutionary strategy to combine these seed functions into more complex functions (e.g., detect and shoot target); and 3) asimulation-based "fitness function" that quantitatively measures the capability of a given agent to perform within a simulated combat scenario. The focus of our product development effort

* Information listed above is at the time of submission. *

US Flag An Official Website of the United States Government