CN109304710A - Mechanical arm precision calibration method based on radial base neural net - Google Patents
Mechanical arm precision calibration method based on radial base neural net Download PDFInfo
- Publication number
- CN109304710A CN109304710A CN201710626910.4A CN201710626910A CN109304710A CN 109304710 A CN109304710 A CN 109304710A CN 201710626910 A CN201710626910 A CN 201710626910A CN 109304710 A CN109304710 A CN 109304710A
- Authority
- CN
- China
- Prior art keywords
- coordinate
- mechanical arm
- point
- tcp
- neural net
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1679—Programme controls characterised by the tasks executed
- B25J9/1692—Calibration of manipulator
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/39—Robotics, robotics to robotics hand
- G05B2219/39024—Calibration of manipulator
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The mechanical arm precision calibration method based on radial base neural net that the invention discloses a kind of.Target ball is installed to the end of mechanical arm to be calibrated, the point using target ball as TCP the following steps are included: erection laser tracker by the present invention；Determine the pose transformation relation of the basis coordinates system of laser tracker coordinate system and mechanical arm；Calculate the error of coordinate between the TCP point theoretical coordinate of multiple reference points and TCP point actual coordinate；Radial base neural net is established, configures that its input be TCP point theoretical coordinate, to export be error of coordinate, and radial base neural net is trained according to the TCP of the multiple reference points obtained before this point theoretical coordinate and error of coordinate；The calibration of mechanical arm precision is carried out using the radial base neural net obtained after training.Mechanical arm precision calibration method based on radial base neural net of the invention, operand is relatively small and is not necessarily to establish the kinematics model of mechanical arm, and then has the advantages that versatility is good, high-efficient.
Description
Technical field
The present invention relates to robot technology, more particularly, to the mechanical arm precision calibration side based on radial base neural net
Method.
Background technique
With the continuous development of science and technology, mechanical arm has been widely used for industry manufacture, military, medical treatment, entertainment garment
The fields such as business, aerospace.Due to the mismachining tolerance of mechanical arm, rigging error, malformation and external disturbance, temperature change
Etc. reasons, cause the actual motion models and theory kinematics model of mechanical arm to have differences, so as to cause manipulator motion
Accuracy decline in the process.Therefore it needs to demarcate the precision of mechanical arm before using mechanical arm, then be mended again
It repays, is allowed to kinematic accuracy raising.The method of currently used mechanical arm calibration needs to repair the kinematics model of mechanical arm
Just, such as " the 6R serial manipulator kinematics parameters identification based on 6 parameter models ".But similar approach in the prior art is most only
Error caused by mechanical arm process and assemble error is considered, and needs to establish kinematics model, there is a problem of computationally intensive.
For this purpose, needing a kind of scaling method of efficient mechanical arm precision.
Summary of the invention
The technical problem to be solved by the present invention is in the prior art the precision of mechanical arm be demarcated to overcome
Method has that computationally intensive, efficiency is not high enough and depends on establishing for kinematics model, proposes a kind of based on radial base
The mechanical arm precision calibration method of neural network.
The present invention is to solve above-mentioned technical problem by following technical proposals:
The mechanical arm precision calibration method based on radial base neural net that the present invention provides a kind of, it is characterized in that,
The following steps are included:
Step 1: setting up laser tracker, target ball is installed to the end of mechanical arm to be calibrated, using target ball as TCP
Point；
Step 2: determining the pose transformation relation of the basis coordinates system of laser tracker coordinate system and mechanical arm；
Step 3: choosing multiple reference points in the working space of mechanical arm, each reference point is sat with TCP point theory
Mark reaches the multiple reference according to control sequence controlled machine arm corresponding with the TCP of each reference point point theoretical coordinate
Point, and coordinate of the corresponding target ball under laser tracker coordinate system is measured using laser tracker, and according to institute's rheme
Appearance transformation relation obtains coordinate of the target ball under the basis coordinates system of mechanical arm by coordinate transform, using as each reference point
TCP point actual coordinate, the coordinate being then calculated between the TCP point theoretical coordinate of each reference point and TCP point actual coordinate miss
Difference；
Step 4: establishing a radial base neural net, the input for configuring the radial base neural net is that TCP point is theoretical
Coordinate, output are error of coordinate, and the TCP point theoretical coordinate and coordinate of the reference point according to obtained in step 1 and three miss
Difference is trained the radial base neural net, obtains radial base neural net after a training；
Step 5: obtaining the theoretical coordinate set of TCP point according to the working space of mechanical arm, and the theory of TCP point is sat
Input of the mark set as radial base neural net after the training obtains the output work of radial base neural net after the training
For the error of coordinate set of TCP point.
The abbreviation of so-called TCP, that is, Tool Center Point, TCP point are meant in tool in the present invention, heart point.
Preferably, the pose transformation relation of the basis coordinates system of laser tracker coordinate system and mechanical arm is laser tracking
The module and carriage transformation matrix of the basis coordinates system of instrument coordinate system and mechanical arm.
Preferably, every control mechanical arm reaches a reference point in step 3, it remain stationary mechanical arm and reaches default
Period, for coordinate of the laser tracker measurement target ball under laser tracker coordinate system.
Preferably, the basic function of the radial base neural net selects Gaussian function.
Preferably, the theoretical coordinate collection of the TCP point in step 5 is combined into the theoretical coordinate complete or collected works of TCP point.
On the basis of common knowledge of the art, above-mentioned each optimum condition, can any combination to get each preferable reality of the present invention
Example.
The positive effect of the present invention is that:
Mechanical arm precision calibration method based on radial base neural net of the invention, operand is relatively small and is not necessarily to
The kinematics model of mechanical arm is established, and then has the advantages that versatility is good, high-efficient.
Detailed description of the invention
Fig. 1 is a preferred embodiment of the present invention mechanical arm precision calibration method, wherein the schematic diagram of used device.
Fig. 2 is a preferred embodiment of the present invention radial base neural net involved in mechanical arm precision calibration method
Schematic illustration.
Specific embodiment
With reference to the accompanying drawings of the specification, further the preferred embodiment of the present invention is described in detail, description below
To be illustrative, not limitation of the present invention, any other similar situation are still fallen among protection scope of the present invention.
In specific descriptions below, the term of directionality, such as "left", "right", "upper", "lower", "front", "rear", etc.,
The direction with reference to described in attached drawing uses.The component of the embodiment of the present invention can be placed in a variety of different directions, directionality
Term is for illustrative purposes and not restrictive.
The mechanical arm precision calibration method based on radial base neural net of a preferred embodiment according to the present invention comprising
Following steps:
Step 1: setting up laser tracker, target ball is installed to the end of mechanical arm to be calibrated, using target ball as TCP
Point；
Step 2: establishing the relative positional relationship of laser tracker and mechanical arm, that is, laser tracker coordinate system OB-
XBYBZBBasis coordinates system O relative to mechanical armA-XAYAZAModule and carriage transformation matrix(with reference to shown in Fig. 1)；
Step 3: choosing multiple reference points in the working space of mechanical arm, each reference point is sat with TCP point theory
Mark reaches the multiple reference according to control sequence controlled machine arm corresponding with the TCP of each reference point point theoretical coordinate
Point, and coordinate of the corresponding target ball under laser tracker coordinate system is measured using laser tracker, and according to institute's rheme
Appearance transformation relation obtains coordinate of the target ball under the basis coordinates system of mechanical arm by coordinate transform, using as each reference point
TCP point actual coordinate, the coordinate being then calculated between the TCP point theoretical coordinate of each reference point and TCP point actual coordinate miss
Difference；
Step 4: establishing a radial base neural net, the input for configuring the radial base neural net is that TCP point is theoretical
Coordinate, output are error of coordinate, and the TCP point theoretical coordinate and coordinate of the reference point according to obtained in step 1 and three miss
Difference is trained the radial base neural net, obtains radial base neural net after a training；
Step 5: obtaining the theoretical coordinate set of TCP point according to the working space of mechanical arm, and the theory of TCP point is sat
Input of the mark set as radial base neural net after the training obtains the output work of radial base neural net after the training
For the error of coordinate set of TCP point.
Refering to what is shown in Fig. 1, according to the mechanical arm precision calibration method of above-mentioned preferred embodiment, used in device substantially
Including mechanical arm A, laser tracker B and target ball C.Wherein, the basis coordinates system of mechanical arm is OA-XAYAZA, laser tracker seat
Mark system is OB-XBYBZB, target ball is mounted on mechanical arm tail end.
Some preferred embodiments according to the present invention specifically can be in the working space of mechanical arm in step 3
Uniformly selection is largely put as a reference point, that is to say the point that TCP point needs to move toAP(AX,AY,AZ) (the i.e. TCP of reference point
Point theoretical coordinate), control mechanical arm reaches the reference point chosen according to the numerical control program woven and is delayed a period of time,
During delay, laser tracker measures target ball coordinate, and obtains target ball in laser tracker coordinate system OB-XBYBZBUnder
CoordinateBP(BX,BY,BZ), by coordinateBP obtains the basis coordinates system O relative to mechanical arm by coordinate transformA-XAYAZAUnder seat
MarkAThe actual coordinate of P ', as TCP point.It can specifically be acquired by following formula:
The then error of coordinate of the theoretical coordinate of TCP point and actual coordinate as a result, are as follows:
ΔAP=AP-AP′
Established in above-mentioned steps four radial base neural net and using Learning Algorithm be trained it is specific can
To be realized using a variety of specific algorithms.Below by with a kind of verified precision in the precision calibration of mechanical arm and efficiency all compared with
The implementation of step 4 is illustrated in good concrete methods of realizing.
Some preferred embodiments according to the present invention establish a radial base neural net in step 4 kind, input and are
The theoretical coordinate of TCP pointAP(AX,AY,AZ), it exports as the error of coordinate Δ of corresponding TCP pointAP(ΔAX, ΔAY, ΔAZ), according to
The data that upper step obtains are trained radial base neural net, obtain a trained radial base neural net.
Wherein, the structure of radial base neural net can be for example shown in Fig. 2.In example shown in Fig. 2, the radial direction base nerve net
Network has 3 inputs, h hidden node, 3 outputs.WhereinAP=(AX,AY,Az)T∈R3For neural network input vector, W ∈ Rh×3
For output matrix, b1, b2, b3For output unit offset, ΔAP=(ΔAX, ΔAY, ΔAz)TFor neural network output vector, Φi
(*) is the activation primitive of i-th of hidden node.In some preferred embodiments of the invention, the base letter of radial base neural net
Gaussian function can be used in number, i.e.,∑ in node layer is exported in Fig. 2 indicates that output layer neuron activates letter
Number, such as linear activation primitive can be used.And ciFor the data center of i-th of hidden node in neural network.
In step 4, the specific implementation step for being related to radial base neural net be can be such that
Specific steps (1) are based on K- means clustering method and solve Basis Function Center ci。
1) netinit: h training sample is randomly selected as cluster centre ci(i=1,2 ..., h), and another k=1.
2) the training sample set of input is grouped by Nearest Neighbor Method: according toAP and ciCenter be between Euclidean distance
It willAP is assigned to each cluster set of input sample | |APj-ci(k) | |, i=1,2 ..., h, j=1, in 2,3.
3) readjust cluster centre: calculating each cluster set | |APj-ci(k) | | the average value of middle training sample, i.e.,
New cluster centre ciIf new cluster centre is no longer changed, obtained ciAs radial base neural net is final
Basis Function Center, otherwise return 2), into next round center solve.
Specific steps (2) solve variances sigmai。
The basic function of radial base neural net of the invention is Gaussian function, therefore variance δiIt can be solved by following formula:
In formula, cmaxMaximum distance between selected center.
Specific steps (3) calculate the weight between hidden layer and output layer.
The connection weight of hidden layer to neuron between output layer can be directly calculated with least square method, be calculated public
Formula is as follows:
W=Φ+ΔAP
Φ+=(ΦTΦ)-1ΦT。
Through above-mentioned steps, may finally be obtained according to the data of the point theoretical coordinate and error of coordinate of TCP obtained in step 3
Radial base neural net to after a training, utilizes the radial base neural net after training, so that it may and then it is directed to mechanical arm
Entire working space or part working space carry out mechanical arm precision calibration.
Although specific embodiments of the present invention have been described above, it will be appreciated by those of skill in the art that these
It is merely illustrative of, protection scope of the present invention is defined by the appended claims.Those skilled in the art is not carrying on the back
Under the premise of from the principle and substance of the present invention, many changes and modifications may be made, but these are changed
Protection scope of the present invention is each fallen with modification.
Claims (5)
1. a kind of mechanical arm precision calibration method based on radial base neural net, which is characterized in that itself the following steps are included:
Step 1: setting up laser tracker, target ball is installed to the end of mechanical arm to be calibrated, the point using target ball as TCP；
Step 2: determining the pose transformation relation of the basis coordinates system of laser tracker coordinate system and mechanical arm；
Step 3: choosing multiple reference points in the working space of mechanical arm, each reference point has TCP point theoretical coordinate, root
The multiple reference point is reached according to control sequence controlled machine arm corresponding with the TCP of each reference point point theoretical coordinate, and
Coordinate of the corresponding target ball under laser tracker coordinate system is measured using laser tracker, and is converted according to the pose
Relationship obtains coordinate of the target ball under the basis coordinates system of mechanical arm by coordinate transform, real using the TCP point as each reference point
Then the error of coordinate between the TCP point theoretical coordinate of each reference point and TCP point actual coordinate is calculated in border coordinate；
Step 4: establishing a radial base neural net, the input of the radial base neural net is configured as TCP point theory seat
Marking, exporting is error of coordinate, and the TCP point theoretical coordinate and error of coordinate of the reference point according to obtained in step 1 and three
The radial base neural net is trained, radial base neural net after a training is obtained；
Step 5: obtain the theoretical coordinate set of TCP point according to the working space of mechanical arm, and by the theoretical coordinate collection of TCP point
Cooperation is the input of radial base neural net after the training, obtains the output conduct of radial base neural net after the training
The error of coordinate set of TCP point.
2. mechanical arm precision calibration method as described in claim 1, which is characterized in that laser tracker coordinate system and mechanical arm
The pose transformation relation of basis coordinates system be that the pose of basis coordinates system of laser tracker coordinate system and mechanical arm converts square
Battle array.
3. mechanical arm precision calibration method as described in claim 1, which is characterized in that in step 3, every control mechanical arm is arrived
Up to a reference point, remain stationary mechanical arm and reach preset time period, for laser tracker measurement target ball laser with
Coordinate under track instrument coordinate system.
4. mechanical arm precision calibration method as described in claim 1, which is characterized in that the base letter of the radial base neural net
Number selects Gaussian function.
5. mechanical arm precision calibration method as described in claim 1, which is characterized in that the reason of the TCP point in step 5
It is the theoretical coordinate complete or collected works of TCP point by coordinate set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710626910.4A CN109304710A (en) | 2017-07-27 | 2017-07-27 | Mechanical arm precision calibration method based on radial base neural net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710626910.4A CN109304710A (en) | 2017-07-27 | 2017-07-27 | Mechanical arm precision calibration method based on radial base neural net |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109304710A true CN109304710A (en) | 2019-02-05 |
Family
ID=65202764
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710626910.4A Pending CN109304710A (en) | 2017-07-27 | 2017-07-27 | Mechanical arm precision calibration method based on radial base neural net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109304710A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111515950A (en) * | 2020-04-28 | 2020-08-11 | 腾讯科技（深圳）有限公司 | Method, device and equipment for determining transformation relation of robot coordinate system and storage medium |
| CN111673739A (en) * | 2020-05-15 | 2020-09-18 | 成都飞机工业（集团）有限责任公司 | Robot pose reachability judgment method based on RBF neural network |
| CN114516055A (en) * | 2022-04-07 | 2022-05-20 | 北京信息科技大学 | Mechanical arm non-shutdown real-time calibration method and device based on binocular vision and deep learning |
| CN116000927A (en) * | 2022-12-29 | 2023-04-25 | 中国工程物理研究院机械制造工艺研究所 | Measuring device and method for spatial position guiding precision of robot vision system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160114486A1 (en) * | 2014-10-27 | 2016-04-28 | Quanta Storage Inc. | Method for automatically recalibrating a robotic arm |
| CN106799745A (en) * | 2017-01-17 | 2017-06-06 | 北京航空航天大学 | A kind of industrial machinery arm precision calibration method based on collocating kriging |
-
2017
- 2017-07-27 CN CN201710626910.4A patent/CN109304710A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160114486A1 (en) * | 2014-10-27 | 2016-04-28 | Quanta Storage Inc. | Method for automatically recalibrating a robotic arm |
| CN106799745A (en) * | 2017-01-17 | 2017-06-06 | 北京航空航天大学 | A kind of industrial machinery arm precision calibration method based on collocating kriging |
Non-Patent Citations (3)
| Title |
|---|
| QIANCHENG ZHU,ET AL.: "Accuracy compensation of a spraying robot based on RBF neural network", 《2016 INTERNATIONAL CONFERENCE ON ADVANCED ROBOTICS AND MECHATRONICS (ICARM)》 * |
| 王小川,等.: "《MATLAB神经网络43个案例分析》", 31 August 2013 * |
| 白云飞,等.: "基于6参数模型的6R串联机器人运动学参数辨识", 《机器人》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111515950A (en) * | 2020-04-28 | 2020-08-11 | 腾讯科技（深圳）有限公司 | Method, device and equipment for determining transformation relation of robot coordinate system and storage medium |
| CN111673739A (en) * | 2020-05-15 | 2020-09-18 | 成都飞机工业（集团）有限责任公司 | Robot pose reachability judgment method based on RBF neural network |
| CN114516055A (en) * | 2022-04-07 | 2022-05-20 | 北京信息科技大学 | Mechanical arm non-shutdown real-time calibration method and device based on binocular vision and deep learning |
| CN114516055B (en) * | 2022-04-07 | 2023-06-06 | 北京信息科技大学 | Real-time calibration method and device for mechanical arm without shutdown based on binocular vision and deep learning |
| CN116000927A (en) * | 2022-12-29 | 2023-04-25 | 中国工程物理研究院机械制造工艺研究所 | Measuring device and method for spatial position guiding precision of robot vision system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109304710A (en) | Mechanical arm precision calibration method based on radial base neural net | |
| CN104057290B (en) | A kind of robotic asssembly method and system of view-based access control model and force-feedback control | |
| CN105259786B (en) | The inertial parameter discrimination method and device of target to be identified | |
| CN102183205A (en) | Method for matching optimal assembly poses of large-sized parts | |
| CN111260649B (en) | Close-range mechanical arm sensing and calibrating method | |
| CN110428464A (en) | Multi-class out-of-order workpiece robot based on deep learning grabs position and orientation estimation method | |
| CN107972027A (en) | The localization method and device of robot, robot | |
| CN111415379A (en) | Three-dimensional point cloud data registration method based on cuckoo optimization | |
| CN110293559A (en) | A kind of installation method of automatic identification positioning alignment | |
| CN110293552A (en) | Mechanical arm control method, device, control equipment and storage medium | |
| CN106408649B (en) | A kind of rule-based body structure evolution design method | |
| CN109130550A (en) | A kind of artificial intelligent control system of electrohydrodynamics spray printing and control method | |
| Edan et al. | Design of an agricultural robot for harvesting melons | |
| CN110962127B (en) | Auxiliary calibration device for tail end pose of mechanical arm and calibration method thereof | |
| CN108646550B (en) | Multi-agent formation method based on behavior selection | |
| CN110070039A (en) | Computer room cabinet and master control borad perception and coordinate measuring method and device based on deep learning | |
| Bao et al. | 3D perception-based collision-free robotic leaf probing for automated indoor plant phenotyping | |
| CN111546379B (en) | Sensor layout method for monitoring operating state of industrial robot | |
| CN109116300A (en) | A kind of limit learning position method based on non-abundant finger print information | |
| Mateo et al. | 3D visual data-driven spatiotemporal deformations for non-rigid object grasping using robot hands | |
| Ninomiya et al. | Automatic calibration of industrial robot and 3D sensors using real-time simulator | |
| CN116500901A (en) | Digital twin-driven man-machine cooperation task planning method under condition of unknown user intention | |
| US20220101477A1 (en) | Visual Interface And Communications Techniques For Use With Robots | |
| CN111906759B (en) | Intelligent moving method of 3D motion mechanism | |
| Li | A Design of Robot System for Rapidly Sorting Express Carton with Mechanical Arm Based on Computer Vision Technology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190205 |