TOYON RESEARCH CORPORATION — Department of Defense SBIR Phase II: N193-A01

Description

In this proposed Phase II SBIR research effort Toyon Research Corporation will apply deep reinforcement learning to the problem of managing decentralized teams of unmanned systems (UMS). Recent applications of Deep Reinforcement Learning techniques in the field of artificial intelligence have resulted in autonomous agents which match or exceed human performance in a variety of decision and control tasks.  In Phase I, Toyon developed the proof-of-concept Modular Autonomy Incubator (MAUI) framework which employed reinforcement learning to train an agent to optimally route multiple UAS assets using only the experience gained by interacting with SLAMEM’s simulated battlespace. The proposed Phase II work plan has been designed to mature the MAUI framework using a Scrum development model which includes three distinct development sprints.  Each sprint is designed to evolve MAUI while using it to train a novel agent for UAS team autonomy. The envisioned Decentralized Unmanned Resource Allocation (DURA) agent will learn to make optimal UAS resource allocation decisions for a large team of loosely connected heterogeneous UAS (i.e., “swarms”). Toyon will deliver MAUI to the Government at the conclusion of each sprint cycle and the DURA agent at the conclusion of Phase II.