Unmanned Air System Search and Localization Guidance Using Reinforcement Learning

Requirments for current and future Unmanned Air Vehicles with longer ight endurances have led to a need for an autonomous soaring capability. This paper investigates aircraft ight path guidance for search and localization of Regions of Interest, consisting of atmospheric phenomena. The problem is posed as an o ine agent learning problem, of localizing atmospheric thermal locations and then guiding an Unmanned Air Vehicle to soar from one to another. Qlearning is used as the learning algorithm. The computational navigation solution used here is a basic grid algorithm that assigns thermal locations and intensities, with the representation being speci ed states, actions, goals, and rewards that are used to accomplish the agent learning. The approach is validated with a simulation of a powered Unmanned Air Vehicle. Results presented in the paper show that the autonomous agent can learn how to navigate to a thermal quickly and e ciently by controlling bank angle, while simultaneously monitoring its inertial position and heading angle.