You are here

A Comprehensive Framework to Develop, Refine, and Validate Learning Agents for Tactical Autonomy

Award Information
Agency: Department of Defense
Branch: Air Force
Contract: FA8650-15-M-6667
Agency Tracking Number: F15A-T14-0236
Amount: $149,633.00
Phase: Phase I
Program: STTR
Solicitation Topic Code: AF15-AT14
Solicitation Number: 2015.1
Solicitation Year: 2015
Award Year: 2015
Award Start Date (Proposal Award Date): 2015-07-29
Award End Date (Contract End Date): 2016-04-29
Small Business Information
7852 Walker Drive Suite 400
Greenbelt, MD 20770
United States
DUNS: 110592016
HUBZone Owned: No
Woman Owned: No
Socially and Economically Disadvantaged: No
Principal Investigator
 Kenneth Center
 (301) 982-6232
Business Contact
 Ella Herz
Phone: (301) 982-6234
Research Institution
 University of Colorado
 Nisar Ahmed
Aerospace Engineering Sciences University of Colorado 429 UCB
Boulder, CO 80309-0429
United States

 (303) 492-0286
 Domestic Nonprofit Research Organization

ABSTRACT: To maintain superiority in the battle domains of space, ground, air and sea, it is natural that the U.S. move toward incorporating autonomous capabilities to enhance/replace the traditional role of humans. Continued advances in processing capability and techniques to employ decision logic and adaptive learning strategies will soon make this vision reality. To deploy any such capability operationally, however, requires a significant level of trust. Building this trust can only be accomplished by employing a comprehensive tool environment enabling high-fidelity battlespace simulations consisting of blue and red actors. Actors will be a mix of humans and agents employing human-like decision logic. Agents participating in these scenarios must be capable of learning from their experiences, and the framework must support the learning process by challenging the actors in ways that maximize learning potential. Orbit Logic and the University of Colorado propose a framework, leveraging a high-heritage distributed simulation environment to host agents. Agent behaviors, in combination with other simulation data sources, will be collected by an agent-learning engine and utilized to improve the actions of the agents both in terms of the breadth of situations that can be recognized and in the effectiveness of its responses to those situations. ; BENEFIT: Completing and commercializing the proposed framework would have nearly immeasurable benefit to all battle domains in the U.S. military purview. Once a basic infrastructure is created to mature and deploy agent-based capabilities, Moores law will apply to the capabilities that can be expected from the agents. Humans are unlikely to ever be completely removed from the decision process, but there is an optimum blend of human and machine strengths that results in the most effective system. This framework will assist in finding that balance. The capabilities of the proposed framework have benefits that apply to virtually any domain that requires human decision making or oversight. While we will be targeting military use cases, the fundamental approaches will be extensible to commercial UAS operations, process control, satellite operations, traffic management, and many more applications in the civil and commercial realm.

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

US Flag An Official Website of the United States Government