Location

Collaborative Visualization Environment (CoVE) Weber Space Science and Technology Building (SST II) and Online:

Sai-Aksharah Sriraman
(Advisor: Prof. Mavris]

will propose a doctoral thesis entitled,

Methodology for Supporting Military Strategy Development at the Operational Level of Warfare using Reinforcement Learning

On

Wednesday, May 28 at 2:00 p.m. ET
Collaborative Visualization Environment (CoVE)

Weber Space Science and Technology Building (SST II)
and

Online: 
https://teams.microsoft.com

Abstract
At the operational level of warfare, commanders must integrate ways (tactics) and means (resources) to achieve objectives. The nature of these objectives requires effort over time at scale with hundreds of sub-objectives, such as in Operation Desert Storm. This involves selecting sub-objectives, deciding when to address each of them, and determining the appropriate resources to use.

 

The United States military has many assets and technologies at its disposal across multiple domains (air, sea, land, etc.), leading to an unfathomable number of courses of action to accomplish an objective. Given the global transition into a Great Power Competition, other powers are rapidly developing and fielding similar technologies, which means that the margin for error on decision-making is slim. Thus, there is a need to rapidly evaluate more strategies to gain more knowledge about addressing a scenario.

 

Wargaming is the established method of evaluating strategies at any level of warfare, where multiple people with varied expertise and experience come together and simulate responses to scenarios. However, this human-based approach limits the speed of the process and inherently introduces subjectivity. Planning in the face of complexity requires more objectivity, which led to the integration of mathematics and computers into the process, or in other words, modeling and simulation (M&S). With M&S techniques, we can simulate scenarios at computer speeds, but the experimentation at the operational level of warfare is still a manual process, both in terms of varying the rule-based strategies created by subject matter experts and analyzing the resulting datasets to improve the strategies.

 

Consequently, many entities are integrating artificial intelligence (AI) methods into the military planning process to explore strategies, primarily at the tactical (most granular) level of warfare. However, there has been little work that applies these methods to the operational level, where the scale and sustainability of forces play a larger role. This thesis proposes a reinforcement learning (RL) methodology to address this gap, adapted from RL applications at the tactical level and from other domains of literature. Reinforcement learning is an AI method where a computer “commander” agent is trained to make decisions within simulated scenarios. The results of the decisions made by this agent can highlight the impacts of different strategies, enabling sensitivity analysis to show how decisions made at different points in time can affect the final outcome.

 

Committee

  • Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)
  • Prof. Kyriakos Vamvoudakis – School of Aerospace Engineering
  • Prof. Jenna Jordan – School of International Affairs
  • Dr. Olivia Pinon-Fischer – School of Aerospace Engineering
  • Dr. Michael Salpukas – Senior Technical Fellow, Raytheon Technologies