A Supervisory Neural Network and Fuzzy Inference System Approach to an Unmanned Aerial Vehicle Tail-Chase Scenario

Abstract

As the capabilities and implementation of multiple Unmanned Aerial Systems (UAS) operations increase, the need to develop reliable, verifiable, and high-performing vehicle control algorithms arises. There are several methods for achieving near optimal navigation control in multiple UAS scenarios. However, many approaches have difficulty identifying boundaries in the state-space where discontinuities in the control signal exist. Neural Networks (NNs) have been shown to be universal approximators of these decision boundaries and can therefore classify the space into discrete regions with differing lower level actions. In this study, a control system for a one-on-one UAS Tail-Chase scenario is developed. An NN was used to define a decision boundary for discrete selection of two different Fuzzy Inference Systems (FISs) for navigation and avoidance control. The parameters for both the NN and FISs are found using a Genetic Algorithm (GA) inside a custom simulation environment. After initial training, uncertainty was included in vehicle movements to improve generalization. The simulation results show that the system was successful in all test cases after adding uncertainty and demonstrate the efficacy of this approach.