Autonomous, Safe and Transparent Reinforcement Learning for Collaborative Air Defense

Purpose and goal

In future combat environment, increasing autonomy of unmanned collaborative combat aircraft brings increased complexity and challenges. These systems must meet performance requirements, be secure, transparent and certifiable. The project aims to develop advanced training and testing methods and tools for design, analysis and validation of secure and interpretable AI agents. To enable training, security analysis and the use of AI agents, a platform for distributed simulation is being established.

Expected effects and result

The developed methods in combination with the simulation platform support academic researchers, actors in industry and Air Force operators in the necessary transformation of the Swedish Air Force to handle the complexity of the future battlespace. An international collaboration will be established for the development of a conceptual platform for distributed simulation.

Planned approach and implementation

The collaboration will unfold in three phases: (1) defining operational scenarios and evaluation criteria through workshops, (2) integrating the simulation environments, and (3) conducting experimental campaigns to evaluate the methods and explore pathways for large-scale demonstrations. Combining inherently transparent RL, explicitly interpretable methods, and simulation environments, the project advances development of safe, trustworthy collaborative autonomy for future air combat systems.