Mobile Robot Control Based on 2D Visual Servoing: A New Approach Combining Neural Network With Variable Structure and Flatness Theory

Abstract

This paper focuses on the 2D visual servo-control of a mobile robot using a neural network (NN) with variable structure. The interaction matrix relating camera movement and changes in visual characteristics requires an estimation phase to determine its parameters as well as a camera calibration phase. It is common in applications related to mobile robotics that the robot model contains uncertainties generated by the sliding phenomenon. We suggest online identification, using NN to avoid this problem. The RBF NN is used to estimate the block formed by the interaction matrix and the reverse robot. Since the number of variables to be estimated is large, this can lead to the use of an excessive number of RBFs. We propose to use a single point of the scene which is sufficient to solve the problem. This problem reduction is possible thanks to flatness theory which allows to reduce the number of NN inputs from 8 inputs (4 image points) -generally used in the literature-to 2 (one image point) only. In order to further reduce the complexity of the proposed algorithm, the number of neurons for each layer and for each iteration is optimized. We use a neural network with variable structure to reach this objective. The very encouraging results obtained validate the proposed approach.

Mobile Robot Control Based on 3D Visual Servoing: A New Approach Combining Pose Estimation by Neural Network and Differential Flatness

“…Finally, S = (x, y) are flat outputs of the considered system. We conclude that the location and the orientation of the robot can be deduced using the coordinates of a single point in the image [10].…”

“…The use of neural networks considerably reduces the complexity of the visual servoing algorithm during its real-time implementation since the system recovers its new position thanks to previously carried out offline learning. According to the theorem announced in [10], we will only consider a single point in the initial image at t and its correspondent in the image taken at t + ∆t, as shown in Figure 3. P(x(t), y(t)): coordinates of P in the image coordinate system at t. P f (xx, yy) : coordinates of P in the image coordinate system at t + 1 (after Translation T and Rotation R) with: xx = x(t + 1) = x(t) + ∆ x (t) and yy = y(t + 1) = y(t) + ∆ y (t).…”

“…Generally, to solve the problem of pose estimation, we need at least three points in the image [1] (six neurons in the input layer), which can generate an excessive number of neurons in the hidden layer. To reduce this number, the concept of differential flatness is used: we proved in [10] that, in the case of a two-degrees-of-freedom mobile robot, the variation in a single point in the picture coordinate space can be used to ensure the robot's position and orientation in relation to the target.…”

An Innovative Collision-Free Image-Based Visual Servoing Method for Mobile Robot Navigation Based on the Path Planning in the Image Plan

“…We showed in [ 35 ] that the coordinates of a single point of the object in the image plane , can act as a flat output for the mobile robot. To simplify the problem, we can consider, in Figure 1 , the coordinate system of the camera confused with the coordinate system of the robot.…”

A New Heuristic for Radius Estimation in RBF Neural Networks

“…The first RBFNN models were proposed more than 30 years ago in [Rumelhart 1986], but they still being actively employed in the industry and academia, for instance in Nonlinear Control ( [Wang et al 2021], [Kaaniche et al 2021]), Sensor Validation ( [Alves et al 2021]), Renewable Energy ( [Barreto et al 2021]), among others applications.…”