In this paper, we discuss both theoretical and implementation issues of a vision based control approach applied to a mobile robot. After having briefly presented a visual servoing framework based on the task function approach, we point out some problems of its application to the case of mobile nonholonomic robots. We will show how using additional degrees of freedom provided by a manipulator arm, allows to overpass these difficulties by introducing redundancy with respect to the task. The second part of the paper deals with the development of an experimental testbed specially designed to rapidly implement and validate reactive vision based tasks. It is constituted by a mobile robot carrying a hand-eye system using a dedicated vision hardware based on VLSI's and DSP 96002's processors, which allows to perform image processing at video rate. Programming aspects are fully taken into account from tasks level specification up to real time implementation by means of powerfull software tools ...

this memory, we will be mostly interested in two of these phases and to the strong interactions binding them: Motion planning and execution. The union of both procedures is used to propose an answer to the general problem of collision-free motion generation.

The image features used in visual servoing or tracking are generally the coordinates of image points. In this paper, we present how it is possible to define more complex visual features based on geometrical primitives such as lines, spheres and cylinders. Using such features enables to realize a large variety of robotics tasks depending on the desired virtual linkage between the camera and the objects in the scene. We then describe a control scheme in closed loop with respect to these image features which is based on the task function approach. This scheme combines the regulation of the selected vision-based task with the minimization of a secondary cost function, such as a trajectory tracking using the robot degrees of freedom not constrained by the vision-based task. We finally present real time experimental results obtained with a camera mounted on the end-effector of a six d.o.f. robot.

This paper deals with vision based control applied to autonomous mobile robots. The general approach in robot vision separates vision from control (static look and move). An alternative way consists to specify the problem in terms of control directly into the sensor frame. This approach seems to be a fruitful way for implementing robotics tasks based on reactivity concept. We introduce a low level task specification by the notion of virtual linkage. This virtual linkage can be expressed by a set of constraints on the motion of a frame linked to the robot with respect to a certain frame associated to the environnment. It seems interesting to apply such a vision based control to a mobile robot. The robot considered here is a two wheels driven nonholonomic cart with a camera mounted on 3 d.o.f manipulator. In this study we consider the whole mechanical system constituted by the cart and its manipulator like a single kinematic chain. So in adding the degrees of freedoom of the arm to the ...

In this paper, we address the problem of generating trajectories of some image features in order to control efficiently a robotic system using an image-based control strategy. First, physically valid C 2 image trajectories which correspond to quasi-optimal 3D camera trajectory are performed. Both self-occlusion avoidance and visibility constraints are taken into account at the task planning level. The good behavior of image-based control when desired and current camera positions are close is then exploited to design an efficient control scheme. Real-time experimental results using a camera mounted on the end effector of a six degree-of-freedom robot confirm the validity of our approach.

This paper describes a real time visual target tracking using the generalized likelihood ratio (GLR) algorithm. We first introduce the visual servoing approach and the application of the task function concept to vision-based tasks. Then, we present a complete control scheme which explicitly enables to pursue a moving object. In order to make the tracking errors as low as possible, we use the GLR test, an algorithm able to detect, estimate and compensate abrupt jumps in target motion. Finally, real-time experimental results using a camera mounted on the end effector of a six-d.o.f. robot are presented. 1 Introduction Visual servoing [5] [6] [8] [10] is now a classical approach to realize various robotics tasks (positioning, grasping, target tracking, etc) in closed loop with respect to visual data. As far as target tracking is concerned, Papanikolopoulos et al. [8] use classical approaches in control theory to track a moving object. However, they consider the object motion as disturb...

With the advent of new visual sensors technology and improved image processing methods, the possibility of relying only on visual information to design closed-loop commands for robot control is not an utopia anymore. This approach brings many advantages such as direct use of image characteristics in control. Instead of intervening at different stages in the control process, robot controllers and image processing modules closely cooperate to achieve a vast collection of various robotics tasks such as assembling mechanical parts or driving a mobile robot. This concept: visual servoing, though fruitful, leads us to face new and unsolved issues. Among them, one is crucial for practical design of such vision-based systems. The information extracted from the images provided by the cameras must be incorporated in a controllable scheme for which a robust command must be found. These control theory considerations can be expressed as: the mapping between the space where controls are performed ...