Graph Reinforcement Learning for Courses of Action Analysis

In land-based operations, the planning staff is faced with the task of assessing the terrain, predicting avenues of approach, and identifying key points for the deployment of forces. To support this process, we developed a graph-based representation of the planning problem. This representation allows us to define different graph-based decision problems, such as identifying key terrain for deploying resources and determining likely courses of action. To solve these problems, we use Graph Reinforcement Learning, a combination of Reinforcement Learning and Graph Neural Networks. This approach allows us to automatically learn a solution strategy for any related problem. To test the effectiveness of our method, we apply it to a specific formulation of a Courses of Action (CoA) analysis, framed as a Network Interdiction problem. We demonstrate the application of Graph Reinforcement Learning in this context and benchmark its performance against traditional Network Interdiction optimization methods. This paper was originally presented at the NATO Science and Technology Organization Symposium (ICMCIS) organized by the Information Systems Technology (IST) Panel, IST-205-RSY - the ICMCIS, held in Koblenz, Germany, 23-24 April 2024.