Digital Night Vision Machine Learning Algorithm for Autonomous Drone Swarm Navigation and Sensing

To address the U.S. Army's demonstrated need for advanced, autonomous Unmanned Aerial System (UAS) navigation and sensing capabilities in very low-light environments, DeepNight proposes an SBIR Phase II effort to develop the Deep Night Vision (DNV) Machine Learning Algorithm for Autonomous Drone Swarm Navigation and Sensing. This effort will enhance Army’s UAS and drone swarm platforms by optimizing the Deep Night Vision (DNV) neural network for real-time navigation and object detection under extremely low-light conditions for UAS-mounted optical sensors. The DNV neural network will enable autonomous navigation, terrain mapping, and target tracking and engagement, significantly improving the operational effectiveness of Army UAS swarms in contested environments by allowing all processing to occur directly on edge devices, reducing object tracking latency and RF emissions. The neural network will specifically implement a digital night vision superresolution algorithm on Army UAS optical sensors, enabling them to sense objects of interest well in excess of their sensing resolution limitations so that they can autonomously identify, track, and engage targets of interest in varied, low-visibility operational settings. As a proof-of-concept to demonstrate the utility and benefits of this capability, DeepNight will also demonstrate early optimization of the DNV UAS algorithm for low-power, constrained compute on UAS chipsets, such as NPU chipsets. By integrating the Deep Night Vision neural network with existing UAS platforms and demonstrating its superior efficacy in autonomous navigation and sensing, DeepNight will provide a scalable, adaptive solution that meets the Army's critical operational needs for enhanced situational awareness and real-time decision-making on UAS platforms in Army operational environments.