Application of Reinforcement Learning to Precision Aerial Delivery System Control in Adverse Wind Conditions

Precision aerial delivery systems (PADS) are a subset of airdropped parachute-leveraging package delivery systems that use autonomous guidance, navigation, and control (GNC) to reach targets with high degrees of accuracy. This technology emerged in the 1990s, and strides have been made since to improve the reliability of traditional physics-based controllers that guide PADS. However, these algorithms still struggle to deliver acceptable performance results when PADS are subjected to austere operating environments, such as those with unpredictable wind. Building on a foundational study in 2022 that used artificial intelligence (AI) and machine learning to improve PADS GNC performance, this study aims to expand on this research and apply reinforcement learning to address the wind problem directly. AI agents enabled to observe simulated wind conditions in real time are trained in three-degree- of- freedom (3DOF) computer simulations and iteratively improved in phases. In the first phase of this study, wind-observing AI agents are developed to navigate PADS in simulation environments with “specific” wind conditions. Then, takeaways from this process are used alongside new training techniques in the second phase of this study to train and test AI agents in environments with “general” wind conditions. The results show that AI agents capable of observing environments’ wind conditions can successfully improve PADS accuracy by one half to one whole order of magnitude, depending on operating conditions.