Robust Neural Network Consensus for Multiagent UASs Based on Weights’ Estimation Error

Abstract

We propose a neural network consensus strategy to solve the leader–follower problem for multiple-rotorcraft unmanned aircraft systems (UASs), where the goal of this work was to improve the learning based on a set of auxiliary variables and first-order filters to obtain the estimation error of the neural weights and to introduce this error information in the update laws. The stability proof was conducted based on Lyapunov’s theory, where we concluded that the formation errors and neural weights’ estimation error were uniformly ultimately bounded. A set of simulation results were conducted in the Gazebo environment to show the efficacy of the novel update laws for the altitude and translational dynamics of a group of UASs. The results showed the benefits and insights into the coordinated control for multiagent systems that considered the weights’ error information compared with the consensus strategy based on classical σ-modification. A comparative study with the performance index ITAE and ITSE showed that the tracking error was reduced by around 45%.

Keywords

• multiagent system
• neural network
• unmanned aircraft systems
• estimation error information