DARPA has released a broad-agency announcement for the Strategic Chaos Engine for Planning, Tactics, Experimentation and Resiliency (SCEPTER) program. SCEPTER will explore the highly complex state action space during military engagements at the speed of a machine, along with discovering novel and unexpected action plans.
SCEPTER’s machine-generated campaign-scale planning and analytics project will develop analytic engines capable of producing machine-generated strategies that can compete with humans in the planning of real warfare. It will achieve a high speed of exploration of action plans through customized abstraction of credible and expert information input models. The best discovered solutions will be validated by researchers in high-fidelity trusted simulators and undergo a comprehensive manual review.
SCEPTER’s first phase will address two key technical focus areas: developing unscripted goal-oriented agents able to discover relevant and interpretable solutions; and managing growth of threats to achieve fast exploration of large-scale military scenarios. Phase two will be a three-year battle planning program.
DARPA plans to spend $39 million US for associated development efforts on the SCEPTER program over the next three years.