Adaptive Visual Servo Control of Robots

Visual servo robot control systems provide feedback on the relative end-effector position of a robot. They offer an interactive positioning mechanism which depends upon extraction and interpretation of visual information from the environment. In this pape

  • PDF / 1,262,741 Bytes
  • 10 Pages / 439.37 x 666.142 pts Page_size
  • 95 Downloads / 250 Views

DOWNLOAD

REPORT


Abstract

Visual servo robot control systems provide feedback on the relative end-effector position of a robot. They offer an interactive positioning mechanism which depends upon extraction and interpretation of visual information from the environment. In this paper, we characterise visual servo systems by the feedback representation mode, position-based or image-based, and the joint control mode, closed-loop or open-loop. The design problems posed by nonlinear and coupling transformations introduced by visual tracking systems are discussed, and a design strategy which utilises adaptive control, direction detection, and a logical control hierarchy is proposed.

Introduction

The interaction of manipulators with sensors has always been an important goal in the development of robotic systems. Such sensor-based systems would have increased functional capabilities as well as flexibility in the execution of tasks. Most practical systems which have exploited this sensor-robot interaction have focused on the development of communications links between existing sensing and robot systems. Such experiments have demonstrated the concept of interactive sensing but have not confronted the basic issues of analysis, design and performance evaluation of such systems. The principal objective of this paper is to describe a formal approach to the analysis of visual servo control structures and the potential use of such tools for the design and evaluation of visual servo systems. A number of factors have delayed the practical development of sensor-based servo systems. Robot positioning systems and robot control systems are difficult to analyse and design in themselves, and there are a large number of practical applications of non-sensing robots in highly structured industrial and other environments. Sensing systems, particularly vision, are often slow relative to manipulator dynamics, and practical applications or machine vision are currently also limited to highly constrained situations. The analytical complexity of both manipulator control and sensory data interpretation make general formulation of the sensor-based control problem challenging. In this paper, we attempt to isolate, as much as possible, the properties of the control problem which arise due to sensory feedback from those issues related to manipulator kinematics and dynamics or specific types of image processing and interpretation. In particular, we focus on the choice of feedback representation

A. Pugh (ed.), Robot Vision © Springer-Verlag Berlin Heidelberg 1983

108 ROBOT VISION

space and distinguish, for the case of vision, between image-based visual feedback which utilises image features as the control parameters, and position-based visual feedback which interprets the image in terms of Euclidean space positions and object geometry and utilises position and orientation parameters for control. Secondly, we have identified two basic classes of control structures. The 'lookand-move' structures utilise inner closed-loop joint control. The 'visual-tracking' structures