Artificial Intelligence and Machine Learning-Based Autonomous Mission Planning for Intelligence, Surveillance, and Reconnaissance (ISR) Missions -- 21

Metron, Inc. and the Maryland Robotics Center (MRC) at the University of Maryland propose to design autonomous, distributed ISR mission planning algorithms deployed on UAV platforms that can replan dynamically based on environmental and mission-specific changes. Agents will negotiate UAV tasks dynamically using AI market-based consensus in which global objectives emerge naturally from UAVs pursuing local goals. Agents will execute UAV routing and control through Bayesian inference and model-based policies that select actions given state based on maximizing value, similar to the role of Q-functions in reinforcement learning. The key challenges are to design algorithms that are distributed, operate in real-time with onboard hardware, address environmental and operational characteristics, and make good, well-informed decisions that are sufficiently interpretable to gain the trust of an operator. We organize our technical system design requirements into three main components: initial mission planning, UAV routing and control, and dynamic replanning.