Research And Design Of Manipula Tor Control System Based On The Visual Feedback

Robot technology has always been the focus of attention. In recent years, more and more people focus on how to make robots get visual ability. Robot system can obtain a lot of data from the external environment through the camera, and identify the target by visual inspection. The purpose of this paper is to design the manipulator control system based on visual feedback which adopts the eye-in-hand architecture, and the system identify the target and the calculated target state of motion by visual processing. It generates the error signal between the end of the robotic arm and objectives in the image space, and plan objectives desired relative speed.Object of this study is RM-501 arm, and to achieve visual tracking system and a manipulator control system. First, according to the geometry manipulator kinematics model is established, and the boundary conditions are set in accordance with the requirements of the tracking task, then this article deduced the relationship between three-dimensional coordinates and joint space coordinates, which preparing for the design of the manipulator control system. Manipulator control system mainly make estimated speed mapping into the joint space, and then make joint motor to track a given angular velocity. Two functional of the visual tracking controller is trajectory planning and BP neural network mapping, that trajectory planning is carried out in the image space and BP neural network will map the result of trajectory planning into the end of the mechanical arm.The focus of this article is to design the visual processing system, which hardware support is the DM6446 dual-core systems. In this paper, the hardware platform is burned into the Linux system, and design image processing program using multi-threaded architecture in the Linux environment. The image processing algorithms is based on the Gaussian pyramid of Lucas-Kanade optical flow. Control program is written and construction in Simulink environment. Finally this paper analyzes the debugging results.