Image Based Visual Servoing: Estimated Image Jacobian by Using Fundamental Matrix VS Analytic Jacobian L. Pari 1 , J.M. Sebastián 1 , A. Traslosheros 1 , and L. Ángel 2 1 Departamento de Automática, Ingeniería Electrónica e Informática Industrial (DISAM) Escuela Técnica Superior de Ingenieros Industriales, Universidad Politécnica de Madrid C/ José Gutiérrez Abascal, 2, 28006 Madrid, Spain Facultad de Ingeniería Electrónica, Universidad Pontificia Bolivariana Km. 7 Via de Piedecuesta, Bucaramanga, Colombia {jsebas,lpari,atraslosheros,langel}@etsii.upm.es Abstract. This paper describes a comparative study of performance between the estimated image Jacobian that come from taking into account the geometry epipolar of a system of two cameras, and the well known analytic image Jaco- bian that is utilized for most applications in visual servoing. Image Based Vis- ual Servoing architecture is used for controlling a 3 d.o.f. articular system using two cameras in eye to hand configuration. Tests in static and dynamic cases were carried out, and showed that the performance of estimated Jacobian by us- ing the properties of the epipolar geometry is such as good and robust against noise as the analytic Jacobian. This fact is considered as an advantage because the estimated Jacobian does not need laborious previous work prior the control task in contrast to the analytic Jacobian does. Keywords: Visual servoing, Jacobian estimation, Fundamental matrix, Interac- tion matrix, robot Jacobian, positioning, tracking. 1 Introduction Visual servoing consists in the use of visual information given by visual sensors (i.e. cameras) to control a robotic system. This kind of control turns out to be very useful in many applications because it allows us to know which objects are present in the scene with high accuracy, as well as their position, orientation and velocity. It makes possible to use robots in new domains where the workspace is not known a priori. Among the existing classifications of visual servoing [4] [10] [11], one of the most known is the way that visual information is used to define the signal error to control the system [2]: Position Based Visual Servoing (PBVS) and the Image Based Visual Ser- voing (IBVS). In PBVS features are extracted from the image and used to reconstruct the 3D position of the target, whereas in IBVS the task is defined in the image plañe di- rectly through image features. In the latter a matrix is defined called the Image Jacobian, which linearly relates changes in image features and changes in Cartesian coordinates or changes in joints (in this case, it is called full-visual-motor Jacobian [1] [5] [11]).