Energy Saving Collaborative Autonomy for Performance Enhancement (ESCAPE)

The burgeoning demands in the military sector, encompassing electronic warfare, advanced sensing, and other multifaceted missions, come head-to-head with the inherent restrictions on the size, weight, and power of energy subsystems in autonomous vehicles. These constraints can critically hamper the vehicle's ability to effectively execute its mission. As today’s Unmanned Air Systems (UAS) capabilities are constantly being pushed, the efficiency and management of energy have become paramount. As the solicitation underscores, there is a pressing need to develop, mature, and implement innovative energy-aware autonomous algorithms that can enhance the electrical power and thermal capabilities of UASs across diverse flight phases. Our Energy Saving Collaborative Autonomy for Performance Enhancement (ESCAPE) solution addresses four key technical challenges related this problem byCollaboration across multiple aircrafts using a Multi-Agent Reinforcement Learning (MARL) solution;Developing trust of ML/AI technology by human operations through using of Explainable AI and smaller sub-task primitives;Incorporation of weather forecasts for pre-flight planning; andDynamic adaptation to changes in the environment and mission during execution.To address these challenges our solution uses a combination of technologies to provide the Air Force with an energy aware collaborative capability for UASs. To address the collaboration challenge, we are proposing to extend our MARL solution to work for multiple mission primitives. In the area of Explainable AI, we are proposing to use a LLM that is capable of decomposing larger mission tasks into smaller primitives that can be executed via collaboration and coordination algorithms. For the weather challenge, we will extend the first-order models that were demonstrated in Phase I for wind currents and fuel consumption to be more realistic and in line with current state-of-the-art. Finally, to demonstrate the dynamic adaptation we will demonstrate our capability in a software environment where we can inject changes to weather forecasts as well as to inject new mission requirements or changes in the UAS mix. The solution progress will be monitored regularly and every 6 months we will perform a development checkpoint that will objectively measure the technical progress of the research. These checkpoints will provide AFRL with an understanding of the impacts of each of the research areas on the overall system performance during the 24-month development.