Dual loop feedbak error learning controller with NARX for mecanum wheels mobile robot

Abstract

A Motorized Adjustable Vertical Platform (MAVeP) is needed by National Space Agency of Malaysia (also known as ANGKASA) at their Satellite Assembly, Integration and Test Centre (AITC), Banting, Selangor. AITC is a clean room used for satellite assessment before it is qualified to be launched into the orbit. Designed as the facility that will provide the similar testing condition as the spacecraft and its payload, AITC is a controlled environment area which having a clean room of class ISO 8. Since it is a confined space for testing, a satellite is carefully transported within the test area as it is very sensitive. Therefore, mecanum wheels are the most suitable wheel for MAVeP mobility mechanism. However, the mecanum wheels come with their common issue like slippage that leads to low accuracy and repeatability of the movement. Each mecanum wheel motor which act as an actuator needs to be properly controlled to avoid overshoot that cause jerk and oscillation that leads to vibration. MAVeP mobility need to transport the satellite with minimum vibration and jerk also achieve positional accuracy and repeatability. This research focuses on the development of dual loop feedback error learning controller with nonlinear autoregressive exogenous neural network (NARX-NN) for MAVeP mobility mechanism. A MAVeP mobility mechanism prototype has been developed and the kinematic model has been derived. Simulations and experiments have been conducted in the linear and diagonal axis. The vibration and jerk issues have been overcame by dual closed loop positioning control system while the slippage problem have been eliminated and improved by using feedback error learning (FEL) control technique which is a dynamic inverse control method that combines simultaneous action of proportional (P) as a feedback controller and NARX as the feedforward controller. NARX learns the inverse dynamic of the MAVeP mobility mechanism in the feedforward controller to improve the response of non-adaptive feedback controller performed by P controller. The experimental result shows that the steady state tracking error is 5%, the maximum overshoot is 0%, the settling time are between 2.9 second to 3.0 second, the RMSE are between 1.83 cm to 2.01 cm and the repeatability are between 98.3% to 100% for all linear movement. Both simulation and experimental results have proven that the proposed controller is successful in controlling the MAVeP mobility mechanism to achieve the desired position accurately and eliminate slippage and jerking.