Two-Layer Neural Network Observer-Based Hierarchical Tracking Control for an Omnidirectional Mobile Robot

Abstract

An approach exploiting the effectiveness of a two-layer neural network for observer design in the tracking control of an omnidirectional mobile robot (OMR) is presented. First, the kinematics and dynamics of the OMR system are formulated to facilitate the observer design. The conditions for ultimate boundedness and the weighted update laws are derived for the neural network observer (NNO) applied to the OMR system. Based on the NNO, velocity information, including lateral, longitudinal, and yaw velocities, which cannot be directly measured, is estimated and used by the tracking controller. A hierarchical control scheme combining a PI controller in the outer loop and an LQR controller in the inner loop is developed based on the NNO to ensure that the OMR follows the desired trajectory. To demonstrate the effectiveness of the proposed method, simulation results are presented under different trajectories, initial estimation errors, and friction coefficients. Comparisons with control methods based on a Luenberger observer and an extended state observer are also provided. The results show that the proposed structure enables efficient and smooth trajectory tracking under different initial observer conditions.