Single-layer neural-network based control of agricultural mobile robot

Abstract

In this study, we propose a novel controller architecture and design for the automatic control of agricultural mobile robots to be used in farms and greenhouses. There are two novelties of this study. The first novelty is a completely new type of controller architecture proposed in which reference inputs and measured outputs are fed separately independent from each other to the controller. The controller architecture currently used in the literature uses only the difference between reference and measurement which is the error signal. The proposed architecture in this study is completely novel in the sense that not only the error information is used in the controller but also the information in reference inputs and information in measured outputs are used separately. This means a completely new type of look to control system by utilizing the information maximally in order to achieve superior performance. This performance boost is shown in the paper where the proposed architecture achieves up to 2000% better performance compared with state-of-the-art controllers. Second, controller architecture is grown to a complex structure from an initially simple PID structure. Using the maximal information comes with the cost of computational complexity to design the controller. The second novelty of growing the controller from initially simple PID equivalent controller tackles this difficulty by making the problem tractable and efficient to compute. This way the proposed novel controller can be designed within minutes in a commercially available laptop computer. The proposed controller is tested on a simulated agricultural mobile robot and results are compared with a previous state-of-the-art optimal controller. It is believed that the proposed architecture will be dominant in future automatic controllers and make current state-of-the-art controllers obsolete. This is because of the full utilization of information in controller design which results in robust disturbance rejection performance.