Abstract
Recent developments in autonomous fighter jets and concepts such as an autonomous wingman are pushing the boundaries for the integration of artificial intelligence (AI) into high-risk military flight operations. Many of these concepts explore how autonomy can accelerate pilot decision-making to reduce cognitive demands, improve data processing, and enhance electronic warfare capabilities. Meeting the demands of the high speed, high risk, and hardware constrained environments typical of military flight has inspired exploration of Neuromorphic Technology. Neuromorphics offers low Size, Weight, and Power (SWaP) and high-speed parallel processing capable of delivering advanced AI abilities. Neuromorphic solutions, which pair biologically inspired hardware and spiking artificial neural networks, are still a nascent technology with many challenges to overcome before they can be fully integrated and flight worthy. One key challenge is demonstrating performance greater than or equal to existing artificial neural networks running on conventional hardware (e.g., Graphics Processing Unit) in an ecologically relevant use case. In this paper, we discuss the trials in developing a closed-loop decision support aid for pilots that combines a spiking neural network (SNN) running on Intel’s Loihi 2 neuromorphic chip with the U.S. Air Force’s Advanced Framework for Simulation, Integration, and Modeling (AFSIM). Specifically, this paper describes architecture and performance of a closed-loop framework supporting use of different versions of a Dense Neural Network (PyTorch on GPU vs SNN on Loihi 2) to generate tactical decision recommendations in an adversarial flight scenario.
