NOU SYSTEMS INC — Department of Defense STTR Phase I: MDA22-T004

Description

State spaces are enormous, the operation is real time, the outcomes are uncertain, the consequences of suboptimal actions are dire. The BMDS is a carefully crafted system with many moving parts. As with all complicated systems, there are many tradeoffs to make when operating them. In a world with finite resources, decisions on how to allocate interceptors during a raid could be the difference between billions of dollars of damage. Moreover, complicated scenarios with decisions that have long term impacts exist in all areas of the military and the commercial sector. Computational decision aids have the potential to assist humans in parsing through the immense web of complicated interactions and side effects – increasing the value our decisions can return. Current state of the art decision making algorithms utilize Deep Reinforcement Learning (DRL). DRL has made incredible breakthroughs in surpassing grandmaster level play in games like Go and Starcraft2. It has also defeated a fighter pilot in a simplified simulation. It is not enough, however, to merely have a DRL warfighter assistant that makes fantastic decisions. It is crucial that this technology can communicate its reasoning behind decisions and integrate into the timeline of how humans execute their decisions. This is not currently achieved by high performing DRL. In explainable reinforcement learning (XRL), most interpretability is currently only delivered by handicapped algorithms on easier tasks. Many tree or graph based algorithms are inherently unable to scale. nou Systems will pioneer interpretability in the DRL space through combining a host of explainability methods into state-of-the-art DRL algorithms. A focus on maintaining an elite level of decision making and setting our algorithms up well to continue to scale is crucial. Furthermore, the temporal nature of the problem is important. Explanations must be crafted to be easily and rapidly understood to enable the warfighter to approve or alter decisions while the engagement is transpiring. Careful integration of the DRL algorithm to be robust to changes in this temporal nature will be developed. A successful DRL warfighter assistant will make excellent decisions, while communicating its reasoning for the decision to the warfighter in a rapidly comprehensible format. In this innovative Phase 1 effort, nou Systems (nSI) will (a) develop a high performant and interpretable DRL algorithm to solve a BMDS relevant task, and (b) construct an integrative dashboard communicating the DRL reasoning and weaving together the DRL and warfighter; resulting optimal decisions approved and trusted by the warfighter. Approved for Public Release | 22-MDA-11339 (13 Dec 22)