Direct Visual Servoing of Planar Manipulators Using Moments of Planar Targets

Abstract

Visual servoing is a control strategy that uses a vision system with one or more cameras to establish the movement of a robotic system (Hutchinson et al., 1996) and emerges as a good alternative to close the control loop between the robot and its environment. Vision is a noncontact method that can relax the setup on an industrial robot and can give more characteristic of autonomy to advanced robots, which operate in adverse ambient or execute service tasks. Depending on the place where the cameras are located, visual servoing is classified by the fixed-camera and camera-in-hand configurations. This chapter addresses the regulation of a planar manipulator by the visual servoing strategy in fixed-camera configuration. To simplify the problem, it is considered that the camera optical axis is perpendicular to the robot motion plane. The specification of the robot motion by the visual servoing strategy must be established trough image features obtained from target objects in the scene or robot workspace. However, most of the developed theory in this area is based on the use of target objects with simple geometry like points, lines, cylinders or spheres (local features); or in more complex target objects but simplifying them to simple geometric objects trough image processing techniques (Collewet & Chaumette, 2000; Benhimane & Malis, 2006). On the other hand, image moments represent global features of an arbitrary target object projected in image plane. The main objective of this chapter is to extend the use of simple target objects toward the use of a more complex target objects in the direct visual servoing of manipulators. Particularly, the target object will be a planar object with arbitrary shape and the image features will be computed trough image moments combinations of this planar target. The direct visual servoing term refers to the class of visual servo-controllers where the visual feedback is converted to joint torques instead to joint or Cartesian velocities; it does mean that the full nonlinear dynamic model of the robot is considered in the control analysis (Hager, 1997). The first explicit solution to the direct visual servoing problem is due to Miyazaki & Masutani (1990). We can find similar works in Espiau et al. (1992) and Kelly et al. (2000) where the problem is approached for a 6 degrees of freedom (6 d.o.f.) manipulator in camera-in-hand configuration. In Kelly (1996); Zergeroglu et al., (1999); Fang et al., (2002); Cheah et al., (2007); and Wang et al. (2008) we can see works that consider the dynamic 22