D.I. Kosmopoulos and C.S. Psomopoulos (Greece)
Image–based visual servoing, Jacobian estimation
A method for estimating through training the feature
Jacobian for image-based closed-loop robot manipulator
control is presented. The resulting Jacobian is
independent of depth information and models the desired
end-effector pose with respect to an image-feature vector.
The estimation is performed locally, in the workspace
volume, where the task is expected to be performed and
therefore is applicable in semi structured environments.
The employment of a robust estimator minimizes the
influence of outliers due to feature extraction noise. The
validity of the method is demonstrated through some
typical experimental results of a robot system with four
cameras mounted on the manipulator.