Comparison of GMDH and Perceptron Controllers for Mobile Robot Obstacle Following/Avoidance with Hardware-in-the-Loop Validation

This paper investigates the effectiveness of using the Group Method of Data Handling (GMDH) and Perceptron neural controllers for a mobile robot obstacle/following avoidance application. The paper evaluates the performance of these controllers in different scenarios, analyzing parameters such as settling time, steady-state error, and overshoot. In addition, we investigate different hardware implementations of the proposed controllers using SoC FPGAs, tailored for small mobile robot platforms, offering high computational performance and low power consumption. To train the neural controllers, four bio-inspired optimization algorithms were used, and hypothesis tests were conducted to select the best neural controller. A Hardware-in-the-Loop (HIL) simulation was conducted in an AMD-Xilinx SoC FPGA Zynq 7020 to attain the best compromise between the controllers’ performance, numerical precision, hardware resources consumption, and power dissipation. The findings underscored the effectiveness of both GMDH and Perceptron controllers in stabilizing the robot amidst disturbances and adeptly navigating obstacle-following and avoidance tasks across various unknown scenarios. However, the Perceptron controller exhibits several advantages in terms of hardware resources and power consumption.