CN-122008244-A - Robot arm grabbing intelligent control method and system based on pose online correction

CN122008244ACN 122008244 ACN122008244 ACN 122008244ACN-122008244-A

Abstract

The invention discloses an intelligent control method and system for mechanical arm grabbing based on pose online correction, and relates to the field of intelligent control of mechanical arms, wherein the method comprises the steps of detecting a mechanical feature matrix through a mechanical sensor at the tail end of the mechanical arm, and predicting drift errors; the method comprises the steps of acquiring binocular images through a binocular vision sensor on a mechanical arm, carrying out registration analysis to obtain registration drift errors and registration drift credibility, verifying to obtain drift consistency parameters, combining the registration drift credibility, marking to obtain a drift prediction training data set, analyzing identification information consistency to obtain identification stability parameters, fusing the prediction drift errors and the registration drift errors to obtain a processed drift prediction training data set, and carrying out mechanical arm correction control and updating training of a drift error predictor. The invention solves the problems of overlarge error, poor correction effect and incapability of adapting to complex industrial environment caused by inaccurate error monitoring of detection equipment in the prior art.

Inventors

YU ZIYONG
Chen Cijian
LI BAINIAN
LUO YONGSHENG
LUO WEI
LI GUOGUANG
WANG PENG
Cen Chaoyu
ZOU JIEXIN
CHEN YUBIN
CHEN HUIMIN
ZHU SIYUAN

Assignees

广东顺畅科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (10)

1. The intelligent control method for the mechanical arm grabbing based on the pose online correction is characterized by comprising the following steps of: Detecting a mechanical characteristic matrix in the grabbing process by a mechanical sensor arranged at the tail end of the mechanical arm, and predicting drift errors by a drift error predictor to obtain predicted drift errors; the binocular vision sensor is configured on the mechanical arm to collect binocular images, and registration analysis is carried out to obtain registration drift errors and registration drift credibility; Verifying the prediction drift error and the registration drift error to obtain drift consistency parameters, and marking a mechanical feature matrix, the prediction drift error and the registration drift error by combining the registration drift reliability to obtain a drift prediction training data set; Analyzing the consistency of the identification information of the drift prediction training data set and the accumulated drift prediction training data set to obtain identification stability parameters, fusing the prediction drift error and the registration drift error to obtain a processed drift prediction training data set, and performing mechanical arm correction control and updating training of a drift error predictor.
2. The intelligent control method for robot arm grabbing based on pose online correction according to claim 1, wherein the mechanical feature matrix is obtained by detection in the grabbing process through a mechanical sensor arranged at the tail end of the robot arm, and the drift error prediction is performed by a drift error predictor to obtain a predicted drift error, comprising: Detecting and obtaining a mechanical feature matrix in the grabbing process through a mechanical sensor arranged at the tail end of the mechanical arm, wherein the mechanical feature matrix comprises mechanical feature values in multiple directions; And inputting the mechanical feature matrix into a drift error predictor, and outputting to obtain a prediction drift error.
3. The robot arm grabbing intelligent control method based on pose online correction according to claim 2, wherein the training step of the drift error predictor comprises the following steps: According to the mechanical arm drift error monitoring data in the historical time, collecting a sample mechanical characteristic matrix set, collecting the scale and the direction of the drift deviation sent by the mechanical arm under different sample mechanical characteristic matrices, and marking to obtain a sample drift error set; Constructing a drift error predictor based on machine learning, wherein the drift error predictor comprises a plurality of groups of initial weights and biases; And training and optimizing the drift error predictor by using the sample mechanical feature matrix set as training input and the sample drift error set as a training supervision label until the test converges, so as to complete training.
4. The intelligent control method for capturing mechanical arm based on pose online correction according to claim 1, wherein the method for capturing the mechanical arm based on pose online correction is characterized by acquiring binocular images through a binocular vision sensor configured on the mechanical arm, performing registration analysis, and obtaining registration drift errors and registration drift credibility, and comprises the following steps: After the mechanical arm is grabbed and returns to a reset state, binocular images are acquired through a binocular vision sensor arranged on the mechanical arm, wherein the binocular images comprise a left image and a right image; Registering the left image and the right image to obtain parallax and registration drift credibility; And identifying the two-dimensional reference coordinates of the reference target in the left image, combining the parallax and the configuration parameters of the binocular sensor, calculating to obtain actual reference coordinates, and calculating the deviation distance between the actual reference coordinates and the preset reference coordinates to obtain registration drift errors.
5. The intelligent control method for capturing by a mechanical arm based on pose online correction according to claim 4, wherein registering the left image and the right image to obtain parallax and obtain registration drift reliability comprises: randomly selecting a first left pixel point in the left image, and extracting a left neighborhood pixel point set of the first left pixel point; selecting a first right pixel point in the right image at random in an iteration way, extracting a right neighborhood pixel point set, calculating the similarity between the right neighborhood pixel point set and a left neighborhood pixel point set, screening a right pixel point with the maximum similarity as the first right pixel point, and calculating the distance between the right pixel point and the first left pixel point to obtain a first pixel point parallax; Continuously calculating to obtain parallax of a plurality of groups of pixel points, and calculating a mean value to obtain parallax; and calculating the discrete parameters of parallax of a plurality of groups of pixel points, and calculating to obtain registration drift credibility.
6. The intelligent control method for capturing by a mechanical arm based on pose online correction according to claim 4, wherein identifying the two-dimensional reference coordinates of the reference target in the left image, combining the configuration parameters of the parallax and the binocular sensor, calculating to obtain an actual reference coordinate, calculating a deviation distance from a preset reference coordinate, and obtaining a registration drift error, comprises: Inputting the left image into a reference target identifier, and identifying and outputting two-dimensional reference coordinates of a reference target, wherein the reference target identifier is constructed based on a convolutional neural network and is obtained by training a sample image set and a sample two-dimensional reference coordinate set; Acquiring main point coordinates, a base line distance and a focal length of the binocular sensor, and calculating to obtain actual three-dimensional coordinates of a reference target as actual reference coordinates by combining the two-dimensional reference coordinates and the parallax; and acquiring a preset three-dimensional coordinate of the reference target, taking the preset three-dimensional coordinate as a preset reference coordinate, and calculating the deviation distance between the preset reference coordinate and the actual reference coordinate to obtain a registration drift error.
7. The intelligent control method for robot arm grabbing based on pose online correction according to claim 1, wherein verifying the prediction drift error and the registration drift error to obtain drift consistency parameters, labeling a mechanical feature matrix, the prediction drift error and the registration drift error in combination with the registration drift reliability to obtain a drift prediction training data set comprises: Calculating the similarity of the prediction drift error and the registration drift error to obtain a drift consistency parameter; calculating to obtain fusion confidence according to the drift consistency parameter and registration drift reliability; and marking the mechanical feature matrix, the prediction drift error and the registration drift error by adopting the fusion confidence coefficient to obtain a drift prediction training data set.
8. The intelligent control method for robot arm grabbing based on pose online correction according to claim 1, wherein analyzing the consistency of the identification information of the drift prediction training data set and the accumulated drift prediction training data set to obtain the identification stability parameter comprises: Acquiring an accumulated drift prediction training data set accumulated in the historical time; And calculating the similarity between the fusion confidence coefficient in the drift prediction training data set and the accumulated fusion confidence coefficient set in the accumulated drift prediction training data set to obtain the identification stability parameter.
9. The intelligent control method for robot arm grabbing based on pose online correction according to claim 1, wherein the method is characterized in that a prediction drift error and a registration drift error are fused to obtain a processed drift prediction training data set, and the robot arm correction control and the updating training of a drift error predictor are performed, and comprises the following steps: configuring registration weights according to the identification stability parameters, and calculating prediction weights; according to the prediction weight and the registration weight, carrying out weighted calculation fusion on the prediction drift error and the registration drift error to obtain fusion drift deviation, and combining fusion confidence to obtain a processed drift prediction training data set; after the accumulated drift prediction training data sets reach the preset quantity, training and updating the drift error predictor; and carrying out grabbing correction control on the mechanical arm according to the fusion drift deviation.
10. The robot arm grabbing intelligent control system based on pose online correction is characterized by being used for implementing the robot arm grabbing intelligent control method based on pose online correction according to any one of claims 1-9, and the system comprises: The drift error prediction module is used for detecting and obtaining a mechanical characteristic matrix in the grabbing process through a mechanical sensor arranged at the tail end of the mechanical arm, and carrying out drift error prediction by adopting a drift error predictor to obtain a predicted drift error; The registration analysis module is used for acquiring binocular images through a binocular vision sensor arranged on the mechanical arm and carrying out registration analysis to obtain registration drift errors and registration drift credibility; the registration reliability labeling module is used for verifying the prediction drift error and the registration drift error to obtain drift consistency parameters, and labeling the mechanical feature matrix, the prediction drift error and the registration drift error by combining the registration drift reliability to obtain a drift prediction training data set; and the correction updating module is used for analyzing the consistency of the identification information of the drift prediction training data set and the accumulated drift prediction training data set to obtain an identification stability parameter, fusing the prediction drift error and the registration drift error to obtain a processed drift prediction training data set, and carrying out mechanical arm correction control and updating training of a drift error predictor.

Description

Robot arm grabbing intelligent control method and system based on pose online correction Technical Field The invention relates to the field of intelligent control of a robot arm, in particular to an intelligent control method and system for robot arm grabbing based on pose online correction. Background Along with the continuous promotion of intelligent manufacturing and production demand, industrial mechanical arms are increasingly widely applied to grabbing operations such as assembly, sorting, feeding and discharging. However, during long-term, continuous operation, the absolute positioning accuracy of the robotic arm system may drift due to a number of factors. Currently, error correction of a mechanical arm mainly depends on-line correction of visual servo or visual guidance, and a camera is used for observing a target or a reference in real time to calculate and compensate pose deviation. However, in a practical complex industrial scene, the reliability of visual perception is highly dependent on ambient illumination and non-occlusion conditions, and when there is reflection or a target is temporarily occluded, the visual information is prone to severe noise. Secondly, the single visual observation value may not be accurate enough for monitoring errors due to errors of an image processing algorithm and the like, and a control strategy cannot be adjusted in a self-adaptive mode. Disclosure of Invention The application provides an intelligent control method and system for mechanical arm grabbing based on pose online correction, and aims to solve the problems that in the prior art, due to inaccurate error monitoring of detection equipment, errors are overlarge in the correction process, so that the correction effect is poor and the method cannot adapt to complex industrial environments. In view of the problems, the application provides an intelligent control method and system for mechanical arm grabbing based on pose online correction. In a first aspect, the application provides an intelligent control method for grabbing a mechanical arm based on pose online correction, which comprises the following steps: Detecting a mechanical characteristic matrix in the grabbing process by a mechanical sensor arranged at the tail end of the mechanical arm, and predicting drift errors by a drift error predictor to obtain predicted drift errors; the binocular vision sensor is configured on the mechanical arm to collect binocular images, and registration analysis is carried out to obtain registration drift errors and registration drift credibility; Verifying the prediction drift error and the registration drift error to obtain drift consistency parameters, and marking a mechanical feature matrix, the prediction drift error and the registration drift error by combining the registration drift reliability to obtain a drift prediction training data set; Analyzing the consistency of the identification information of the drift prediction training data set and the accumulated drift prediction training data set to obtain identification stability parameters, fusing the prediction drift error and the registration drift error to obtain a processed drift prediction training data set, and performing mechanical arm correction control and updating training of a drift error predictor. In a second aspect, the invention provides an intelligent control system for robot arm gripping based on pose online correction, the system comprising: The drift error prediction module is used for detecting and obtaining a mechanical characteristic matrix in the grabbing process through a mechanical sensor arranged at the tail end of the mechanical arm, and carrying out drift error prediction by adopting a drift error predictor to obtain a predicted drift error; The registration analysis module is used for acquiring binocular images through a binocular vision sensor arranged on the mechanical arm and carrying out registration analysis to obtain registration drift errors and registration drift credibility; the registration reliability labeling module is used for verifying the prediction drift error and the registration drift error to obtain drift consistency parameters, and labeling the mechanical feature matrix, the prediction drift error and the registration drift error by combining the registration drift reliability to obtain a drift prediction training data set; and the correction updating module is used for analyzing the consistency of the identification information of the drift prediction training data set and the accumulated drift prediction training data set to obtain an identification stability parameter, fusing the prediction drift error and the registration drift error to obtain a processed drift prediction training data set, and carrying out mechanical arm correction control and updating training of a drift error predictor. One or more technical schemes provided by the application have at least the following technical effects or advantages: Acc