Image-based visual servoing for mobile robots using neural networks and fuzzy-evolutionary methods

Presents an image based visual servoing method, which plans the navigation trajectory in the image plane for non-holonomic mobile robots. It does not require camera calibration nor coordinate transformation from image space to workspace. The robot navigates along a preplanned workspace trajectory while at the same time it learns, using neural networks, a desired or reference landmark trajectory in the image space. Each point of the reference trajectory is basically what is seen of the target landmark at each control cycle. After this network training, the robot compares the current landmark with not only the target landmark but also the reference landmark along the way. Using a fuzzy logic to control this error the mobile robot navigates through the learned trajectory in the workspace. The fuzzy logic is optimized by an evolutionary algorithm to meet a user-defined objective. The proposed algorithm has been verified via simulation.