| Assembly task is the main link of industrial production and manufacturing.The quality of assembly often directly determines the quality of the final product.Positioning is an important basis for assembly,and its accuracy and efficiency directly affect the final quality of assembly.The use of industrial robots can replace humans to complete assembly work more accurately and efficiently,and in some harsh environments,the use of robots to achieve automatic positioning is the current development trend.At present,industrial robots mainly rely on offline programming,teaching,or using laser trackers for positioning when completing assembly tasks.Both offline programming and teaching programming have problems such as poor versatility,lack of perception,and accuracy affected by clamping accuracy;the use of laser trackers is poor in flexibility,expensive,and signal continuity is easily affected.Robot positioning based on visual servoing provides a feasible way of thinking due to its flexibility,continuous visual information,strong versatility,and low cost.Compared with position-based visual servoing,image-based visual servoing has great advantages in robot positioning tasks: it does not require complex feature extraction and3 D reconstruction,and has low requirements for computing power;for robot model errors,calibration errors,etc.The interference factor is more robust.Therefore,this thesis mainly focuses on image-based visual servoing,and conducts in-depth research on it from three perspectives: error convergence structure,adaptive gain,image characteristics and error measurement,and improves the efficiency and robustness of visual servo positioning in assembly.(1)The experimental platform of the visual servo positioning system is built and its software and hardware components,camera model and robot model are introduced.In view of the two problems encountered in the actual deployment of the visual servo positioning system: the calculation of large amounts of data requires high computer performance,and the unstructured and cluttered environment in the actual assembly scene affects the effect of visual servoing.Deployment strategy including pose replication.(2)Propose a new error convergence structure,and design the controller.Quantitative comparative experiments were carried out with the second-order controller and the classic first-order controller,and the comparative analysis was carried out from the three aspects of convergence rate,robust performance to depth value and speed fluctuation.Verify the feasibility of the proposed controller and its superior performance in terms of convergence speed and robustness to depth.(3)The non-linear differential-tracker is used to smooth the discrete speed signals,which makes up for the lack of speed fluctuations of the designed new controller and improves the performance of the controller.It is proposed to use PD controller instead of static gain,so that the system can approach the desired position with higher efficiency even at the stage of small error;the gain decay based on error steady state is proposed to ensure that the desired position can remain stable.And it is verified on the simulation-physical experiment platform of the visual servo positioning system that the speed smoothing processing and adaptive gain can effectively suppress the speed jitter.(4)Using the histogram feature as the global feature of the image,it is proposed to use the cosine distance of the histogram feature as the error metric between the real-time acquired image and the expected image,deriving the corresponding interaction matrix and using the image division method to achieve the six-dimensional camera Speed control.Compared with using Mahalanobis distance as the error metric,the efficiency of this method is verified by simulation.The effectiveness of the proposed method in flat and non-planar scenes is verified on the Rokae robot experimental platform,and the robustness to depth and partial occlusion is further verified. |
PDF Full Text Request