Control in Robotics and Automation

60

CHAPTER 2 / OBSERVER-BASED VISUAL SERVOING

are presented. One is a nonlinear model-based controller and the other is a linearized version

of the nonlinear controller. The nonlinear controller is applied to a direct drive robot for

which nonlinear dynamics cannot be neglected. The linearized controller is used for a general

six-degree-of-freedom geared robot.

The stability of the observer-based control system is presented and the effectiveness of the

observer is verified by experiments on a two-link direct drive robot and a PUMA 560.

1 INTRODUCTION

Visual information on tasks and environments is essential for robots to execute flexible and

autonomous tasks. A typical task of autonomous manipulation is to track a moving object

with a robot hand based on the information from a vision sensor. To carry out this task, the

vision sensor must be incorporated in the feedback loop. Figure 2.1 shows an example of a

visual feedback task. A camera is mounted on the robot hand and it captures images of the

object. An image is considered as a two-dimensional array of gray-level signals whose size is

typically 512 x 512 pixels. If the gray levels from all pixels are considered as the measurement

signal, the system will not be manageable because the size of the measurement vector is larger

than 200,000 and each element has a nonlinear correlation with its neighbors. Therefore,

preprocessing of the raw image is necessary. Usually, image features of the object are

extracted by preprocessing. A few examples of research based on the stochastic models of

two-dimensional observations are found (e.g. [1]), but most visual servoing schemes use the

features of the image as an observation. An image feature is any structural feature that can

be extracted from an image (e.g., an edge or a corner). Typically, an image feature will

correspond to the projection of a physical feature of the object [27]. The robot is controlled

on the basis of the image features, and further image processing (e.g., image understanding

or recognition) is omitted.

There are two approaches in visual feedback control:

position based

and

feature based

[46]. With position-based schemes, the object position and orientation relative to the camera

are computed by using photogrammetric [12,47,49], stereo [1, 38, 39], or "depth from

motion" techniques [28, 35]. Because the position of the object is available as the output of

the image processing part, a conventional position controller can be used to control the

manipulator. However, geometric model of the object is required and the camera-robot

FIGURE 2.1

Visual tracking.

Object

Image

Robot .......

s ~

Camera

/