Non-linear model predictive control for visual servoing systems incorporating iterative linear quadratic Gaussian

Abstract

This study proposes a modified non-linear model predictive control (NMPC) to deal with the point stabilisation for the image-based visual servoing (IBVS) of a wheeled mobile robot by using the iterative linear quadratic Gaussian (iLQG) algorithm. Firstly, an IBVS non-linear system is modelled by taking the image coordinates and angle into account simultaneously. By setting a loss function, a model predictive control minimisation problem with an input constraint is formulated. Then, an iLQGbased NMPC strategy is proposed to obtain the optimal solution where a dual-gradient descent is used to deal with the input constraint and a specific Lorentzian ρ-function is added into the loss function to reduce the static error. Finally, numerical simulations and several comparisons are demonstrated to show the effectiveness of the proposed algorithm.