SCALABLE MULTI-AGENT COLLABORATION FOR RADAR TASKS

Abstract

We examine the challenge of multi-agent command and control (C2) scenarioswhere autonomous agents accomplish objectives by observing their environment, deciding optimal actions to meet the objective, and then act upon their environment. The most common method for these agents to observe their environment is through the use of radar. The domain of cognitive radar follows a similar perception-action cycle and accomplishes radar resource management (RRM) tasks. Modern C2 challenges with multiple autonomous agents using radar presents a unique challenge to both domains. Multi-agent autonomy is often formulated without the uncertainty and complexity of higher-fidelity radar and radar task methods often optimize singular systems as opposed to coalitions of agents. The integration of radar task methods with multi-agent systems presents a unique challenge that requires maximization of collaborative utility instead of singular performance. This work examines C2 challenges where agents have increased uncertainty and reduced resources but can and should coordinate actions. We designed a simulation environment for C2 challenges with medium fidelity radar detection simulations. Within this environment, we present agent action methods in the fields of machine learning and reinforcement learning to perform with low resources and increased uncertainty. Novel to this work is the use of game theory and meta-cognition to coordinate multiple agents for increase utility in the radar tasks of detection, track confirmation, and optimal surveillance. In order to meet the needs of the C2 domain, special attention is paid to methods that are scalable and modular.