COMBAT: Constructive Machine-learning Battles with Adversary Tactics

Project Lead:  Dr. Paul Zablocky                                                            

Sponsoring Organization:  DARPA

Website:  https://www.darpa.mil/program/constructive-machine-learning-battles-with-adversary-tactics

Project Synopsis:  Army force-on-force war games employ Red Team (adversary) operators who rely upon long-established techniques, tactics, procedures, and conventional force employment against their Blue opponent (U.S. Forces). COMBAT aims to develop artificial intelligence (AI) algorithms to generate models of Red Force brigade behaviors that challenge and adapt to Blue Forces in simulation experiments. This effort will investigate how an adversary may evolve from current to future tactics based on Blue Force actions and responses. If successful, COMBAT will rapidly develop multiple feasible Red Force courses of action, identify optimal solutions, and provide reasoning to support recommendations. In future war games, COMBAT will deliver an AI Red Force capable of reacting in ways not easily replicated by human role-players, challenging Blue role-players and stimulating the development of new Blue tactics.