Hardware-in-the-Loop Simulation of Autonomous System using Neural Network in Small Computer Platform

Unmanned Aerial Vehicle (UAV) also called unmanned aircraft is often used for various purposes such as photography, carrying goods, to industrial monitoring. Most of the control schemes for the UAV such as quadcopter is using PID based controller because of low computational cost and simplicity, but this type of controller is unable to compensate for nonlinear disturbances such as harsh wind or frame vibration. This can be solved by using neural network control but there are several constraints in UAV application, such as weight limitation, power limitation, even compatibility of the system that can be solved by using small computer platform. Continuing on previous research for low level quadcopter control based on neural network, the next step is to implement this control in Hardware-in-the-loop format safely by using Pressure Process Rig data that had the same nonlinear characteristic. This paper explores Hardware-in-the-loop simulation development with several small computer platform in the neural network control application. In addition, variation between several hardware and TensorFlow version is carried out to determine the effect of each variation. This paper shows how the various hardware and even different version of programming library can affect the performance of the neural network model computation.