A visual robotic arm grasping method and device applied to parametric parts

Abstract

A visual manipulator grasping method and device applied to parametric parts, the method comprising: S1, obtaining scene point clouds of parametric parts, removing background information to obtain target point clouds; S2, inputting target point clouds into parameterized points The cloud deep neural network maps the target point cloud to the target point in the feature vector space according to the descriptor mapping function generated by the parameterized point cloud deep neural network; For the manifold of the part family, by calculating the feature point closest to the target point on the surface of the manifold, the specific parameter value of the target object and the corresponding target template are obtained; S4. According to the target template, the 6D pose of the parameterized part is obtained through the alignment algorithm Information; S5. Transmitting the 6D pose information of the parameterized part to the control system of the robotic arm to realize the grasping of the parameterized part. The invention is suitable for grabbing parts of different part families, saves a lot of calculation time, and has strong robustness and versatility.

Description

technical field [0001] The invention relates to visual grasping, in particular to a visual manipulator grasping method and device applied to parametric parts. Background technique [0002] my country's demand for robots is increasing day by day, and robots are being widely used in the field of parts grabbing. However, most of the robots currently being used in the society lack intelligence, and it is difficult to adapt to the production mode of many varieties and small batches and the unstructured industrial environment in industry. [0003] Visual grasping refers to guiding the robot to grasp the target object through the visual system. CN201511005603 discloses a visual robot grasping method based on product information labels. The image processing control device obtains the position and size information of the object, and finally controls the robot to realize the grasping action. CN201810034599 discloses a robot grasping method based on depth vision, using an edge detect...

AI Technical Summary

Problems solved by technology

Since the grasping methods commonly used in the current technology are mostly aimed at the single object grasping scene with a fixed template, it is necessary to build all the part templates in the entire part family in advance, and the workload is large.

Moreover, different part families have different feature distributions, and traditional visual recognition and grasping methods are difficult to apply to object models of different part families.

Images