Deep Reinforcement-learning for Autonomously Controlled Optimized Systems (DRACOs)

Toyon would like to propose a novel discover and tracking of HVT with swarms of small Unmanned Aerial Vehicles (sUAS) controlled by Artificial Intelligence agents. Given a covert mission behind enemy lines, we envision a set of sUAS which are available from companies, such as AeroVironment’s Switchblades, could be stationed in discrete launchers near a High Value Target (HVT) or multiple HVT sites. Once triggered that a HVT has left the area given trusted intel, the Deep Reinforcement Learning (Deep RL) agent will make a split second decision based upon the type of HVT (person walking, truck driving, etc.) on the number of assets to be assigned to a particular HVT. The agent will also consider whether or not to launch additional assets for proper coverage given assets that are already in the air and consider other mission critical tasks, such as contested or hostile environments. Toyon plans to utilize our in-house Modular Autonomy Incubator to train a Deep RL agent (Toyon’s Digital Twin) tuned to the dynamic real-time environment described above. Toyon envisions placing all of the technologies onto a small edge computing device such as an Nvidia Xavier. This Xavier board is optimized for Deep Reinforcement learning tasks and with its small SWAP footprint is ideal for being onboard an sUAS.