Abstract

The end positioning process of the grasping manipulator is affected by environmental interference , Resulting in poor positioning accuracy and stability

• Establish the end error model of the grab manipulator by fusing multi-sensor data
• The target detection model determines the target location
• Genetic algorithm calculates the rotation variable compensation value of each joint at the end of the grasping manipulator

key word

1. Kinematic model ;
2. End pose error ;
3. Biplane constraint error ;
4. object detection ;
5. Genetic algorithm (ga) ;

0 introduction

With the improvement of science and Technology , It also puts forward higher requirements for the end positioning of the grab manipulator .
It is not only required to grasp the end of the manipulator Accurate positioning , It is also required to have Good confidence .

Previous studies

1 Design of grab manipulator model fused with multi-sensor data

1.1 Kinematic model of manipulator fused with multi-sensor data

1） Establish the kinematics model of the manipulator

The coordinate system is established with the end of the grasping manipulator fused with multi-sensor data as the target
The change matrix between adjacent coordinate systems is determined by 4 Description of sensor capture parameters
$$Z=(x,y)=\left[\begin{array}{cccc}\alpha & -\alpha & 0& y\\ \alpha ,y& x,y& -y& -e,y\\ x,y& x,-y& y& e,y\\ 0& 0& 0& 1\end{array}\right]$$
$Z$ It represents the change matrix ;$x$、$y$ It describes the coordinate point at the end of the grasping manipulator ;$\alpha$ Represents the rotation angle between coordinate systems ;$e$ Represents the moving distance between coordinate points .

2） The pose matrix of the end of the grasping manipulator

$$\{\begin{array}{l}U=\left[\begin{array}{cc}Z& U\\ 0& 1\end{array}\right]\\ U=U_1,U_2,U_3,U_4\end{array}$$
$U$ Represents the pose matrix .

3） Error description

Multisensor data contains Jamming data , As a result, the coordinate system in the kinematic model of the gripper manipulator cannot Completely parallel or vertical

1.2 Construction of error model of multi-sensor data of terminal pose

The parameter error data collected by multiple sensors , Combined with pose matrix , Build the end pose error model of the manipulator

$W$ Is the parameter error .
fitting (3) After simplification, we can get , Identify all parameters in the error matrix as shown in (4) Shown

$K$ Is the transformation parameter .

1.3 Biplane constraint error model

When positioning the end of the grab manipulator by fusing multi-sensor data , structure Biplane constraint error model To eliminate the edge data in the data , Make the positioning result more accurate .

1. Usage (5) Calculate the actual position of the end of the grasping manipulator , Then calculate the Jacobian matrix of the actual position through the joint angle .
   
2. According to the coordinate points on any constraint plane, the kinematic parameter error at the end of the grasping manipulator is calculated , Put the error into the controller that fuses multi-sensor data , Obtain the actual coordinate point at the end of the manipulator through the updated formula , Compare with the position point on another constraint plane .
   
3. Collect on the biplane constraint error model $C+3$ A little bit , And three consecutive points cannot be in the same straight line , The final biplane constraint error model is obtained by the following formula , Complete the elimination of the end edge data of the grasping manipulator .
   

1.4 Target detection model

The target detection model is mainly composed of neurons in convolution layer and junction layer

1. Convolution layer is responsible for Convolution characteristic graph Lieutenant general Convolution kernel By convolution , To obtain the eigenvector . Then the eigenvector is transferred to the connection layer .
2. The connection layer is responsible for Predict the location of the area to be selected , And judge whether the current area to be selected is in the target area . When the center point of the moving window corresponds to the target area , This location can be regarded as a target location , Through the processing of connection layer , The regression boundary of all target positions can be obtained .
3. Linear regression algorithm is adopted , Implement Micro adjustment , So that the target can be positioned more accurately , Calculate the translation and zoom parameters through the following formula .
4. The feature vector is transmitted to the terminal actuator that fuses multi-sensor data , Initialize parameters , Repeat steps $1$, Until the feature extraction of the image at the end of the grasping manipulator is completed .

2 Robot arm end positioning based on Genetic Algorithm

Due to kinematic parameters Random change
There is a certain error in the pose parameters of the manipulator
Proposed in Genetic algorithm (ga) Take corrective measures for the joint variables of the manipulator on the basis of
Greatly improve the end positioning of the manipulator accuracy

1. Sort out Multiple groups Manipulator data as a reference target , choice Any group of pose information
   According to the three-dimensional translation model, the actual pose data of the sensor terminal actuator is drawn
   according to Genetic variable algorithm , Calculate the ideal compensation value of each joint variable
   Put the corrected ideal compensation value into the multi-sensor terminal control system , Get the actual pose error information .
2. Design genetic algorithm parameters .
   Pose error correction solution
   Take these error values as the parameters of genetic algorithm , The implementation of Copy 、 variation Wait for the operation , Generate a new generation of parameters , Until the error between the actual position and the target position is small enough .
3. The realization of genetic algorithm .
   For the theoretical pose data processed by the multi-sensor terminal actuator , Calculate the corresponding grasping joint variables .
   Take the grasping joint variable as the control quantity , use NDI The system measures the error between the actual posture and the theoretical posture .

3 Experiment and analysis

1） End positioning efficiency

Compare the positioning time used by the three methods , The longer the positioning time , It shows that the lower the positioning efficiency , contrary , The shorter the positioning time , It shows that the higher the positioning efficiency .

2） Positioning error

3） Degree of confidence

Confidence refers to the distribution of the observed value of the end positioning of the grab manipulator , It is a normal distribution with ideal observations as the mean , It represents the error distribution function between the positioning result and the ideal result . The higher the confidence value , It shows that the more accurate the positioning is ; The lower the confidence value , It indicates that the larger the positioning deviation .

4 Conclusion