Deep reinforcement learning based unManned ARiel VEhicLes (D-MARVEL)

The ability to quickly adapt to the multitude of dynamic mission objectives and environments that UAV solutions are subjected to is a key differentiator between autonomous UAV solutions. Furthermore, it is often the case that an operator must team with the autonomous UAV in an effort to maximize the efficacy of the team as a whole. Restrictions such these render more traditional solutions of ranging based mapping and solely vision-based algorithms inert as the former can be time and power consumptive and the latter may be restricted to a sole objective. Moreover, neither directly account for the need to team with a human operator. Thus, ANDRO proposes Deep reinforcement learning based unmanned Ariel VEhicles (D-MARVEL), a dynamically weighted mutli-objective deep reinforcement learning solution providing autonomy, adaptability, and human machine teaming for UAVs. D-MARVEL’s learning framework, developed using the open source architecture, provides a modular and reusable solution to various UAV platforms and other UAS alike. Additionally, ANDRO proposes the use of a natural user interface within the D-MARVEL learning framework, which allows for communication with the UAV without a direct communication link, and further emphasizes D-MARVEL’s applicability to other UAS platforms.