Design And Research Of Royal Jelly Collecting Robot System Based On Monocular Vision

Royal jelly is a popular nutritional health products, but its acquisition oftraditional manual process is still in an inefficient way. Because of thedispersibility and small size of queen larvae in a beehive, it’s difficult forbeekeepers to identify them, especially for long time tremendousconcentration. In order to slove the problem, the visual module based on therobot body was increased. And also its3D modeling and kinematicssimulation were built to verify the feasibility of the system function. Finally,the robot error modeling and its compensation method were established. Andit indicated the effectivity of the compensation method from the simulationexperiment.Firstly, the paper introduced the traditional way of royal jelly acquisitionand the existing problems as well as domestic and foreign research on itbriefly. Considering the existing problems, the royal jelly collecting robotsystem based on monocular vision is pointed out. And then the primarycoverage of this paper is introduced.Secondly, the task flow of the robot execution is designed based on theartificial acquisition process and the function of the system is analized.According to its function, the system can be divided into visual recognitionpart and machine implementation part. The robot structure is also described inthe paper briefly.Thirdly, the paper introduced the design of visual processing module andits camera model in detail. The key features of this part are image processing, edge extracting, shape recognition and calculation of the target points. Andfinally the points are converted from image space to the world coordinatethrough the camera model.Fourthly, in order to uncouple visual processing module dependence onthe robot, the robot simulation module is designed according to concurrentengineering (CE). As an independent software, the module can receive fromand respond to visual processing module with custom communicationprotocol. In order to make the simulation results reflect the movement of therobot as realistically as possible, the kinematics part of the two modules readthe same configuration file. Besides the simulation module and the robotreceive the same data from the visual processing moduel.Finally, the pose error of the robot end-effector was analyzed. Thegeometric error model of the robot was also built for analysis of the absolutepositioning error of the robot. Next, the error compensation method wasproposed based on the error model and it’s proved to be feasible through thesimulation experiment.