Sub-optimal Perceptron Controller Using WOA Algorithm for Farming Watering Mobile Robot

Abstract

Artificial neural networks (ANNs) are intricate mathematical models, drawing inspiration from the biological nervous system, and offering intelligence alongside nonparametric capabilities. The efficacy of ANNs heavily relies on their learning process. Heuristic search algorithms have been proposed in literature as reliable options for optimizing neural networks utilizing the Perceptron architecture. This paper presents an suboptimal neural network Perceptron training method employing the Whale Optimization Algorithm for the application of farming watering mobile robots. To achieve this, a fitness function based on system responses such as overshoot, settling Time, and steady-state Error was developed. The sub-optimal Perceptron controller was then implemented using MATLAB, CoppeliaSim software and the EVA robot, resulting in a 66.17% overshoot, a 4.7% steady-state error, and a settling time of 3.05 seconds. Moreover, the sub-optimal Perceptron controller was mapped onto a System on Chip FPGA AMD-Xilinx Zynq 7020, resulting in a hardware resources consumption of 1,870 (3.52%) Lookup table, 2 (0.9%) Digital Signal Processing blocks, 1,019 (0.96%) Flip-Flops and a power consumtion of 0.005 W. Finally, Hardware-in-the-Loop validation was performed using CoppeliaSim, Python, Universal Direct Memory Access, and the EVA robot. The results showed an overshoot of 94.83%, a settling time of 2 seconds, and a steady-state error of 4.5%.