CN110039537A - A kind of automatic measure on line multi joint motion planing method neural network based - Google Patents
A kind of automatic measure on line multi joint motion planing method neural network based Download PDFInfo
- Publication number
- CN110039537A CN110039537A CN201910198188.8A CN201910198188A CN110039537A CN 110039537 A CN110039537 A CN 110039537A CN 201910198188 A CN201910198188 A CN 201910198188A CN 110039537 A CN110039537 A CN 110039537A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- input
- output
- hidden layer
- 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.)
- Granted
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
- B25J9/1656—Programme controls characterised by programming, planning systems for manipulators
- B25J9/1664—Programme controls characterised by programming, planning systems for manipulators characterised by motion, path, trajectory planning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a kind of automatic measure on line multi joint motion planing methods neural network based to realize the real-time self study and control of non-linear pahtfinder hard using the best approximation capability of the overall situation of neural network.Trajectory planning control is carried out since multi-joint is synchronous, it is with the time in complicated non-linear relation, the present invention is by establishing the neural network model of the time correlation sequence of each joint position, realize the automatic measure on line of multi-joint, real-time control for intelligent machine arm, the difficulty of numerical solution is greatly reduced, improves operation efficiency, and there is real-time self-learning capability.
Description
Technical field
The present invention relates to a kind of automatic measure on line multi joint motion planing methods neural network based, for learning by oneself online
The motion planning for realizing multi-joint mechanical arm and control are practised, intelligent robot trajectory planning and control field are belonged to, it is particularly suitable
In realization mechanical arm to the discovery of new trajectory path, real-time self study and control.
Background technique
With the development of technology, intelligent robot has obtained extensive concern and application, it comprises mechanical structure, drives
The multi-disciplinary highly integrated equipment such as dynamic, control, the mechanical arm of intelligent robot generally comprise multiple driving joints, track control
System determines precision, service and the application of mechanical arm.Manipulator motion TRAJECTORY CONTROL is chiefly used in three-dimensional space, common at present
The trajectory planning of state space parsing and software motion the Realization of Simulation mechanical arm multi-joint.But the mechanical arm of three-dimensional space
Workspace calculation is related to a variety of sampling and solves numerical method, and multi-joint is synchronous to carry out trajectory planning control, and it is in the time
Complicated non-linear relation, solution procedure is cumbersome, difficulty is big and efficiency is lower, has lag characteristic, is difficult to realize mechanical arm
Self study real-time motion planning.
Summary of the invention
It is an object of the invention to overcome the above-mentioned deficiency of the prior art, a kind of online self-study neural network based is provided
Multi joint motion planing method is practised, the difficulty of numerical solution is greatly reduced, improves operation efficiency, and there is real-time self study
Ability.
Above-mentioned purpose of the invention is achieved by following technical solution:
A kind of automatic measure on line multi joint motion planing method neural network based, includes the following steps:
(1) it by the sensor acquisition trajectories relevant information on mechanical arm, and records, the data of preceding n-hour acquisition are denoted as
φ (t-1), φ (t-2) ... φ (t-N), t are current time, and N is the natural number greater than 1, and the data of each moment acquisition are used
Vector representation, wherein including the track-related information in each joint of mechanical arm;
(2) the self study stage is determined whether, if entering step (3) in the self study stage, if not in self study rank
Section, enters step (4);
(3) using the track-related information φ (t-1) of preceding n-hour, φ (t-2) ... φ (t-N) as input, to work as
Preceding moment track-related information φ (t) establishes neural network model as output, carries out neural network instruction according to the collected data
Practice, updates neural network model;After the complete running track data for needing to learn of acquisition, the training of neural network is completed,
Jump out the self study stage;
(4) by φ (t), φ (t-1), φ (t-2) ... φ (t-N-1) as neural network model input, utilize step
(3) neural network model updated in calculates the output φ (t+1) of neural network model, realizes PREDICTIVE CONTROL.
In the step (1), track-related information is joint position angle information, moment information or posture information.
In the step (3), neural network model includes input layer, hidden layer and output layer;The output quantity of input layer is
The input quantity of hidden layer;The output quantity of hidden layer is the input quantity of output layer.
Hidden layer node is 3-5 node.
In the step (3), when establishing neural network model, it is thus necessary to determine that the excitation function between input layer and output layer
Relationship.
Specific step is as follows for the excitation function relationship for determining between input layer and output layer:
S1: the input quantity I1 of hidden layer is calculatediWith the input quantity φ (t-1) of input layer, φ (t-2) ... φ's (t-N)
Relationship;
S2: the input quantity I1 of hidden layer is calculatediWith the output quantity O1 of hidden layeriRelationship;
S3: the output quantity of output layer is calculatedWith the output quantity O1 of hidden layeriRelationship;
S4: the relationship of input quantity of the input quantity and input layer of above-mentioned hidden layer, the input quantity of hidden layer and hidden layer
The relationship of the relationship of output quantity and the output quantity of output layer and the output quantity of hidden layer is between input layer and output layer
Excitation function relationship.
In the step S1, the input quantity I1 of hidden layer is calculated using following formulaiWith the input quantity φ (t- of input layer
1), φ (t-2) ... the relationship of φ (t-N):
In formula, m1 is the number of hidden layer node;I is i-th of hidden layer node;J is j-th of input layer;θ1jFor
The threshold value of j-th of input layer;w1i,jFor the connection weight parameter of i-th of hidden layer node and j-th of input layer.
In the step S2, the input quantity I1 of hidden layer is calculated using following formulaiWith the output quantity O1 of hidden layeriPass
System;
In formula, a is tilt parameters.
In the step S3, the output quantity of output layer is calculated using following formulaWith the output quantity O1 of hidden layeriPass
System:
Enable the threshold θ 1 of input layerjIt is 0, then:
In formula, viFor the connection weight parameter for exporting node layer and hidden layer node;
M2 is output layer node number.
M2 takes 1.
The invention has the following advantages over the prior art:
(1) present invention realizes the real-time self-study of non-linear pahtfinder hard using the best approximation capability of the overall situation of neural network
It practises and control, multi-joint synchronizes progress trajectory planning control, difficulty in computation is small, and computational efficiency is high.
(2) neural network model that the present invention is established using the physical relation of the time correlation sequence of each joint position,
The prediction for realizing joint angles, to realize the compensation of time delay in control process.
Detailed description of the invention
Fig. 1 is flow chart of the present invention;
Fig. 2 is neural network model schematic diagram.
Specific embodiment
The present invention is described in further detail in the following with reference to the drawings and specific embodiments:
Neural network method can ignore the specific physical parameter of process or system, real by the study to training sample
Complex nonlinear mapping between existing input and output, and there is preferable generalization ability, there is global best approximation capability, training
Method is quickly easy, and local optimum problem is not present.Using the track-related informations such as neural network joint position angle with
Functional relationship between the track-related information of preceding n-hour, can predict track-related information.
Specifically, as shown in Figure 1, automatic measure on line multi joint motion planing method of the present invention is realized by following steps:
(1) pass through sensor acquisition trajectories relevant information (joint position angle information, moment information or the appearance on mechanical arm
State information), and record.The data of preceding n-hour acquisition are denoted as φ (t-1), φ (t-2) ... φ (t-N), and t is current time, N
For the natural number greater than 1, the data vector representation of each moment acquisition, wherein the track comprising each joint of mechanical arm is related
Information.
(2) the self study stage is determined whether, if entering step (3) neural net model establishing process in the self study stage,
If entering step (4) PREDICTIVE CONTROL process not in the self study stage.
(3) defeated as neural network using φ (t-1), φ (t-2) ... φ (t-N) as neural network input φ (t)
Out, neural network model is established, carries out neural metwork training according to the collected data, updates neural network model;When having acquired
After the data of whole service track, the training of neural network is completed, jumps out the self study stage.Neural network model is as shown in Figure 2.
(4) φ (t), φ (t-1), φ (t-2) ... φ (t-N-1) are inputted as neural network, using in step (3)
The neural network model updated calculates neural network output φ (t+1), realizes PREDICTIVE CONTROL.
Further, in step (3), neural network model includes input layer, hidden layer and output layer;The output of input layer
Amount is the input quantity of hidden layer;The output quantity of hidden layer is the input quantity of output layer;Hidden layer node is 3-5 node.
When establishing neural network model, it is thus necessary to determine that the excitation function relationship between input layer and output layer, i.e. hidden layer
Excitation function, the specific steps are as follows:
S1: the input quantity I1 of hidden layer is calculated using formula (1)iWith input quantity φ (t-1), the φ (t- of input layer
2) ... the relationship of φ (t-N);
In formula,
M1 is the number of hidden layer node;
I is i-th of hidden layer node;
J is j-th of input layer;
θ1jFor the threshold value of j-th of input layer;
w1i,jFor the connection weight parameter of i-th of hidden layer node and j-th of input layer.
S2: the input quantity I1 of hidden layer is calculated using formula (2)iWith the output quantity O1 of hidden layeriRelationship;
In formula, a is tilt parameters.
S3: the output quantity of output layer is calculatedWith the output quantity O1 of hidden layeriRelationship;
Calculation method is as follows:
Enable the threshold θ 1 of input layerjIt is 0, then:
In formula, viFor the connection weight parameter for exporting node layer and hidden layer node;
M2 is output layer node number, takes 1.
S4: the relationship of input quantity of the input quantity and input layer of above-mentioned hidden layer, the input quantity of hidden layer and hidden layer
The relationship of the relationship of output quantity and the output quantity of output layer and the output quantity of hidden layer is between input layer and output layer
Excitation function relationship.
The present invention using neural network the best approximation capability of the overall situation, realize non-linear pahtfinder hard real-time self study and
Control.Trajectory planning control is carried out since multi-joint is synchronous, with the time in complicated non-linear relation, the present invention passes through foundation
The neural network model of the time correlation sequence of each joint position, realizes the automatic measure on line of multi-joint, is used for intelligent machine
The real-time control of arm greatly reduces the difficulty of numerical solution, improves operation efficiency, and have real-time self-learning capability.
The content that description in the present invention is not described in detail belongs to the well-known technique of those skilled in the art.
Claims (10)
1. a kind of automatic measure on line multi joint motion planing method neural network based, characterized by the following steps:
(1) it by the sensor acquisition trajectories relevant information on mechanical arm, and records, the data of preceding n-hour acquisition are denoted as φ (t-
1), φ (t-2) ... φ (t-N), t are current time, and N is the natural number greater than 1, the data vector table of each moment acquisition
Show, wherein including the track-related information in each joint of mechanical arm;
(2) the self study stage is determined whether, if entering step (3) in the self study stage, if not in the self study stage, into
Enter step (4);
(3) using the track-related information φ (t-1) of preceding n-hour, φ (t-2) ... φ (t-N) as input, with it is current when
Track-related information φ (t) is carved as output, neural network model is established, carries out neural metwork training according to the collected data,
Update neural network model;After the complete running track data for needing to learn of acquisition, the training of neural network is completed, is jumped out
The self study stage;
(4) by φ (t), φ (t-1), φ (t-2) ... φ (t-N-1) as neural network model input, utilize step (3)
The middle neural network model updated calculates the output φ (t+1) of neural network model, realizes PREDICTIVE CONTROL.
2. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 1, special
Sign is: in the step (1), track-related information is joint position angle information, moment information or posture information.
3. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 1, special
Sign is: in the step (3), neural network model includes input layer, hidden layer and output layer;The output quantity of input layer is hidden
Input quantity containing layer;The output quantity of hidden layer is the input quantity of output layer.
4. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 3, special
Sign is: hidden layer node is 3-5 node.
5. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 3, special
Sign is: in the step (3), when establishing neural network model, it is thus necessary to determine that the excitation function between input layer and output layer
Relationship.
6. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 5, special
Sign is: specific step is as follows for the excitation function relationship for determining between input layer and output layer:
S1: the input quantity I1 of hidden layer is calculatediWith the input quantity φ (t-1) of input layer, φ (t-2) ... the relationship of φ (t-N);
S2: the input quantity I1 of hidden layer is calculatediWith the output quantity O1 of hidden layeriRelationship;
S3: the output quantity of output layer is calculatedWith the output quantity O1 of hidden layeriRelationship;
S4: the output of the relationship of input quantity, the input quantity of hidden layer and hidden layer of the input quantity and input layer of above-mentioned hidden layer
The relationship of the relationship of amount and the output quantity of output layer and the output quantity of hidden layer is the excitation between input layer and output layer
Functional relation.
7. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 6, special
Sign is: in the step S1, the input quantity I1 of hidden layer is calculated using following formulaiWith the input quantity φ (t-1) of input layer,
φ (t-2) ... the relationship of φ (t-N):
In formula, m1 is the number of hidden layer node;I is i-th of hidden layer node;J is j-th of input layer;θ1jIt is j-th
The threshold value of input layer;w1i,jFor the connection weight parameter of i-th of hidden layer node and j-th of input layer.
8. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 7, special
Sign is: in the step S2, the input quantity I1 of hidden layer is calculated using following formulaiWith the output quantity O1 of hidden layeriPass
System;
In formula, a is tilt parameters.
9. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 8, special
Sign is: in the step S3, the output quantity of output layer is calculated using following formulaWith the output quantity O1 of hidden layeriPass
System:
Enable the threshold θ 1 of input layerjIt is 0, then:
In formula, viFor the connection weight parameter for exporting node layer and hidden layer node;
M2 is output layer node number.
10. a kind of automatic measure on line multi joint motion planing method neural network based according to claim 9, special
Sign is: m2 takes 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910198188.8A CN110039537B (en) | 2019-03-15 | 2019-03-15 | Online self-learning multi-joint motion planning method based on neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910198188.8A CN110039537B (en) | 2019-03-15 | 2019-03-15 | Online self-learning multi-joint motion planning method based on neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110039537A true CN110039537A (en) | 2019-07-23 |
CN110039537B CN110039537B (en) | 2021-07-13 |
Family
ID=67273732
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910198188.8A Active CN110039537B (en) | 2019-03-15 | 2019-03-15 | Online self-learning multi-joint motion planning method based on neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110039537B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110826701A (en) * | 2019-11-15 | 2020-02-21 | 北京邮电大学 | Method for carrying out system identification on two-degree-of-freedom flexible leg based on BP neural network algorithm |
CN111590581A (en) * | 2020-05-26 | 2020-08-28 | 珠海格力智能装备有限公司 | Positioning compensation method and device for robot |
CN113515123A (en) * | 2021-06-25 | 2021-10-19 | 北京精密机电控制设备研究所 | Robot real-time path planning method based on improved RRT algorithm |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101825870A (en) * | 2010-05-18 | 2010-09-08 | 浙江浙大中控信息技术有限公司 | Method and system for controlling supply quantity of water-treatment flocculating agent |
CN106874914A (en) * | 2017-01-12 | 2017-06-20 | 华南理工大学 | A kind of industrial machinery arm visual spatial attention method based on depth convolutional neural networks |
CN106914897A (en) * | 2017-03-31 | 2017-07-04 | 长安大学 | Inverse Solution For Manipulator Kinematics method based on RBF neural |
WO2018022718A1 (en) * | 2016-07-26 | 2018-02-01 | University Of Connecticut | Skill transfer from a person to a robot |
CN108038541A (en) * | 2017-12-07 | 2018-05-15 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and the computer-readable medium that CTR is estimated |
CN108453737A (en) * | 2018-03-23 | 2018-08-28 | 南京工业大学 | Robot action track acquisition system and method based on neural network |
CN108478222A (en) * | 2017-12-21 | 2018-09-04 | 北京精密机电控制设备研究所 | A kind of joint angle error compensation experimental provision and method based on neural network |
CN109291052A (en) * | 2018-10-26 | 2019-02-01 | 山东师范大学 | A kind of massaging manipulator training method based on deeply study |
-
2019
- 2019-03-15 CN CN201910198188.8A patent/CN110039537B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101825870A (en) * | 2010-05-18 | 2010-09-08 | 浙江浙大中控信息技术有限公司 | Method and system for controlling supply quantity of water-treatment flocculating agent |
WO2018022718A1 (en) * | 2016-07-26 | 2018-02-01 | University Of Connecticut | Skill transfer from a person to a robot |
CN106874914A (en) * | 2017-01-12 | 2017-06-20 | 华南理工大学 | A kind of industrial machinery arm visual spatial attention method based on depth convolutional neural networks |
CN106914897A (en) * | 2017-03-31 | 2017-07-04 | 长安大学 | Inverse Solution For Manipulator Kinematics method based on RBF neural |
CN108038541A (en) * | 2017-12-07 | 2018-05-15 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and the computer-readable medium that CTR is estimated |
CN108478222A (en) * | 2017-12-21 | 2018-09-04 | 北京精密机电控制设备研究所 | A kind of joint angle error compensation experimental provision and method based on neural network |
CN108453737A (en) * | 2018-03-23 | 2018-08-28 | 南京工业大学 | Robot action track acquisition system and method based on neural network |
CN109291052A (en) * | 2018-10-26 | 2019-02-01 | 山东师范大学 | A kind of massaging manipulator training method based on deeply study |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110826701A (en) * | 2019-11-15 | 2020-02-21 | 北京邮电大学 | Method for carrying out system identification on two-degree-of-freedom flexible leg based on BP neural network algorithm |
CN111590581A (en) * | 2020-05-26 | 2020-08-28 | 珠海格力智能装备有限公司 | Positioning compensation method and device for robot |
CN111590581B (en) * | 2020-05-26 | 2021-10-22 | 珠海格力智能装备有限公司 | Positioning compensation method and device for robot |
CN113515123A (en) * | 2021-06-25 | 2021-10-19 | 北京精密机电控制设备研究所 | Robot real-time path planning method based on improved RRT algorithm |
CN113515123B (en) * | 2021-06-25 | 2024-04-12 | 北京精密机电控制设备研究所 | Robot real-time path planning method based on improved RRT algorithm |
Also Published As
Publication number | Publication date |
---|---|
CN110039537B (en) | 2021-07-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wen et al. | Path planning for active SLAM based on deep reinforcement learning under unknown environments | |
JP6869562B2 (en) | A method of tracking an object using a CNN including a tracking network, and a device using it {METHOD FOR TRACKING OBJECT BY USING CONVOLUTIONAL NEURAL NETWORK INCLUDING TRACKING NETWORK AND COMPUTING | |
CN110379247B (en) | Multitask and multi-role ROV (remote operated vehicle) collaborative training simulation system and method | |
CN110039537A (en) | A kind of automatic measure on line multi joint motion planing method neural network based | |
CN104808590B (en) | Mobile robot visual servo control method based on key frame strategy | |
CN107862970B (en) | Teaching quality evaluation model for turnover classroom | |
CN107861508A (en) | A kind of mobile robot local motion method and device for planning | |
WO2022105635A1 (en) | Robot motor-skill learning method and system | |
CN106022380A (en) | Individual identity identification method based on deep learning | |
CN109464803A (en) | Virtual objects controlled, model training method, device, storage medium and equipment | |
CN109102525A (en) | A kind of mobile robot follow-up control method based on the estimation of adaptive pose | |
CN109782600A (en) | A method of autonomous mobile robot navigation system is established by virtual environment | |
CN110472738A (en) | A kind of unmanned boat Real Time Obstacle Avoiding algorithm based on deeply study | |
CN104408760A (en) | Binocular-vision-based high-precision virtual assembling system algorithm | |
CN110328668B (en) | Mechanical arm path planning method based on speed smooth deterministic strategy gradient | |
CN110481536A (en) | A kind of control method and equipment applied to hybrid vehicle | |
CN111857107A (en) | Auxiliary mobile robot navigation control system and method based on learning component library | |
CN110154024A (en) | A kind of assembly control method based on shot and long term Memory Neural Networks incremental model | |
CN109919302A (en) | A kind of neural network training method and device | |
CN113759901A (en) | Mobile robot autonomous obstacle avoidance method based on deep reinforcement learning | |
CN112102681A (en) | Standard motor train unit driving simulation training system and method based on self-adaptive strategy | |
CN112857370A (en) | Robot map-free navigation method based on time sequence information modeling | |
CN108415445A (en) | A kind of submarine navigation device positioning robust Optimal methods based on switch constraint | |
Wang et al. | A two-layer trajectory tracking control scheme of manipulator based on elm-smc for autonomous robotic vehicle | |
CN109764876A (en) | The multi-modal fusion localization method of unmanned platform |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |