Visual path following by an omnidirectional mobile robot using 2d visual servoing

Abstract

This paper presents the simulation of path following by an autonomous omnidirectional mobile robot with an onboard monocular camera. The control of the mobile robot for performing the aforementioned task is done with a 2D visual servoing by considering point features. Artificial landmarks are supposed to be placed in the robot's navigation environment. We performed this simulation in two ways. In the first way, we considered only the 3D positions of the visual features and the model of the monocular camera in order to determine the images of the visual features needed for the mobile robot control. In the second way, the 3D navigation environment of the robot (walls, floor, landmarks … etc.) is modeled by a virtual reality model. Thanks to MATLAB's “Virtual Reality” toolbox, it is possible to move the viewpoint and capture the watched scenes. The visuals features are extracted by processing the captured images. The first simulation is a simplification of the second since the image processing step for extracting the visual features is omitted. However, the second simulation is closer to the practical problem of mobile robot visual control. The first simulation that is admissible in many cases does not allow easy modeling of occultation problems. Also, we focused in this paper on the optimization of the execution time of the control program; by proposing a solution based on the position error threshold of the landmarks image with respect to the reference image.