The Counter-UAS Reinforcement learning Environment (The CURE)

The Counter-UAS Reinforcement learning Environment (The CURE) is a flexible, architecture-agnostic reinforcement learning (RL) training, evaluation, and demonstration environment alongside a set of state-of-the-art benchmark agents. The CURE is designed to allow researchers and RL experts to design, train, and test groundbreaking deep RL agents to avoid or defeat counter-UAS (C-UAS) challenges while completing complex, multi-task goals. The CURE also allows training and demonstration of multi-agent control algorithms to allow for management of multiple UASs with optional human interaction. The CURE neatly packages the necessary back-end infrastructure for producing a sufficient variety of high-fidelity challenge scenarios at the quantity and speed required for deep RL training. In addition, The CURE computes, tracks, and compiles key evaluation metrics over the training history, enabling advanced training curricula that are responsive to performance convergence and breakpoints. The CURE’s benchmark agents, trained and evaluated entirely within The CURE’s framework, are state-of-the art architectures designed to provide a starting point of performance and reasoning for new agents trained and evaluated on The CURE. This platform will enable the next generation of autonomous reasoning algorithms for swarms of UASs to defeat or evade C-UAS tactics in dynamic, multi-task operational environments.