CN114970305A - Prediction method for magnetic particle distribution of magnetic control software robot - Google Patents
Prediction method for magnetic particle distribution of magnetic control software robot Download PDFInfo
- Publication number
- CN114970305A CN114970305A CN202111553563.XA CN202111553563A CN114970305A CN 114970305 A CN114970305 A CN 114970305A CN 202111553563 A CN202111553563 A CN 202111553563A CN 114970305 A CN114970305 A CN 114970305A
- Authority
- CN
- China
- Prior art keywords
- magnetic
- motion state
- distribution
- training
- magnetic field
- 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
- 238000009826 distribution Methods 0.000 title claims abstract description 77
- 239000006249 magnetic particle Substances 0.000 title claims abstract description 75
- 238000000034 method Methods 0.000 title claims abstract description 42
- 230000008569 process Effects 0.000 claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 15
- 238000013461 design Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 55
- 238000012360 testing method Methods 0.000 claims description 26
- 230000004044 response Effects 0.000 abstract description 7
- 238000004364 calculation method Methods 0.000 abstract description 6
- 238000004088 simulation Methods 0.000 abstract description 6
- 238000012827 research and development Methods 0.000 abstract description 4
- 238000013507 mapping Methods 0.000 abstract description 3
- 238000005457 optimization Methods 0.000 abstract description 3
- 238000005452 bending Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010408 sweeping Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J19/00—Accessories fitted to manipulators, e.g. for monitoring, for viewing; Safety devices combined with or specially adapted for use in connection with manipulators
- B25J19/007—Means or methods for designing or fabricating manipulators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/25—Design optimisation, verification or simulation using particle-based methods
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medical Informatics (AREA)
- Computer Graphics (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Manipulator (AREA)
Abstract
The invention discloses a method for predicting magnetic particle distribution of a magnetic control software robot, which comprises the following steps: observing the motion state of the magnetic control soft robot to be detected under a specific magnetic field, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state; and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the distribution of the magnetic particles in the magnetic control software robot. The method searches the mapping relation between the response and the model structure from the simulation data, avoids the traditional mathematical physical calculation method, omits the processes of modeling, parameter scanning optimization calculation and the like in the traditional design method, improves the research and development efficiency, and avoids overhigh cost caused by blind design.
Description
Technical Field
The disclosure relates to a magnetic control soft robot, in particular to a method for predicting the distribution of magnetic particles of a magnetic control soft robot.
Background
Magnetic controlled robots are one of the most interesting issues in the field of robot research today, and magnetic driving methods have been considered and proven to be a promising technology due to the ability to generate forces and torques contactlessly through magnetic fields. The key technology of the magnetic control soft robot is preparation technology and control technology. From the aspect of motion, for a magnetic control soft robot, the distribution mode of magnetic particles in the magnetic control soft robot and the distribution of an external magnetic field directly influence the motion mode of the magnetic control soft robot, including speed, direction, action and the like, and a specific internal magnetic particle distribution and magnetic field distribution model is designed, so that how the magnetic control soft robot can move under the drive of a specific magnetic field according to the expected motion mode is a key problem to be solved.
Disclosure of Invention
In view of the deficiencies in the prior art, the present disclosure provides a method for predicting magnetic particle distribution of a magnetic control soft robot, which uses a corresponding relationship between magnetic particle distribution in the robot and a motion state of the robot or a corresponding relationship between a magnetic field of the robot and the motion state of the robot as training sample input, and predicts the magnetic particle distribution in the magnetic control soft robot and the magnetic field of the magnetic control soft robot through a deep neural network, so as to improve research and development efficiency and avoid overhigh cost caused by blind design.
In order to achieve the above purpose, the present disclosure provides the following technical solutions:
a magnetic control software robot magnetic particle distribution prediction method comprises the following steps:
s100: observing the motion state of the magnetic control soft robot to be detected under a specific magnetic field, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state;
s200: and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the distribution of the magnetic particles in the magnetic control soft robot.
Preferably, the predictive model includes, and is not limited to, DNN, RNN or GAN networks.
Preferably, the training process of the prediction model is represented as follows:
s201: under a specific magnetic field, designing magnetic particle distribution in the magnetic control soft robot in different modes, and recording the motion state of the robot corresponding to each design to obtain a first magnetic particle distribution-motion state corresponding relation data set;
s202: respectively carrying out digital processing on the distribution and motion state of all the magnetic particles in the first magnetic particle distribution-motion state corresponding relation data set to obtain a second magnetic particle distribution-motion state corresponding relation data set;
s203: and dividing the second magnetic particle distribution-motion state corresponding relation data set into a training set and a testing set, training the prediction model through the training set and testing the trained prediction model through the testing set, finishing model training if a testing result meets a preset accuracy rate, and otherwise, optimizing network parameters and training configuration parameters of the prediction model until the training requirements are met.
Preferably, in step S202, the digitizing the magnetic particle distribution includes: and expressing the part containing the magnetic particles by using a coordinate sequence, and meshing the part containing the magnetic particles to obtain coordinates of each node, wherein each coordinate expresses a cluster of agglomerated magnetic particles.
Preferably, in step S202, the digitizing the motion state includes: and expressing the magnetic particles corresponding to the final motion state by using a coordinate sequence, and performing grid division on the magnetic particles to obtain coordinates of each node, wherein each coordinate expresses a cluster of agglomerated magnetic particles.
The present disclosure also provides a magnetic control software robot magnetic field distribution prediction method, including the following steps:
s1000: observing the motion state of the magnetic control soft robot to be detected under the distribution of specific magnetic particles, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state;
s2000: and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the magnetic field of the magnetic control software robot.
Preferably, the predictive model includes, and is not limited to, DNN, RNN or GAN networks.
Preferably, the training process of the prediction model is represented as follows:
s2001: under the distribution of specific magnetic particles, changing the size and direction of a magnetic field outside the magnetically controlled soft robot, and recording the motion state of the robot corresponding to the magnetic field, so as to obtain a first magnetic field distribution-motion state corresponding relation data set;
s2002: respectively carrying out digital processing on all the magnetic field distributions and the motion states in the first magnetic field distribution-motion state corresponding relation data set to obtain a second magnetic field distribution-motion state corresponding relation data set;
s2003: and dividing the second magnetic field distribution-motion state corresponding relation data set into a training set and a testing set, training the prediction model through the training set, testing the trained prediction model through the testing set, finishing the model training if the test result meets the preset accuracy, and otherwise, optimizing the network parameters and the training configuration parameters of the prediction model until the training requirements are met.
Compared with the prior art, the beneficial effect that this disclosure brought does: the method can search the mapping relation between the response and the model structure from the simulation data, can avoid the traditional mathematical physical calculation method, can obtain the model structure corresponding to the response at the output end only by inputting the expected response at the input end of the network, omits the processes of modeling, parameter sweeping optimization calculation and the like in the traditional design method, improves the research and development efficiency, and avoids overhigh cost caused by blind design.
Drawings
Fig. 1 is a flowchart of a method for predicting magnetic field distribution of a magnetically controlled soft robot according to an embodiment of the present disclosure;
fig. 2 is a schematic simulation structure diagram of a helmholtz coil three-dimensional model according to an embodiment of the present disclosure;
fig. 3(a) to 3(c) are distribution patterns of magnetic particles inside a robot according to an embodiment of the present disclosure;
FIG. 4 is a schematic diagram of magnetic particle pillar meshing provided by an embodiment of the present disclosure;
fig. 5(a) to 5(c) are schematic diagrams of a motion state of a robot provided in an embodiment of the present disclosure;
fig. 6 is a schematic structural diagram of a GAN network according to an embodiment of the present disclosure;
fig. 7 is a schematic structural diagram of a DNN network provided in an embodiment of the present disclosure.
Detailed Description
Specific embodiments of the present disclosure will be described in detail below with reference to fig. 1 to 7. While specific embodiments of the disclosure are shown in the drawings, it should be understood that the disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
It should be noted that certain terms are used throughout the description and claims to refer to particular components. As one skilled in the art will appreciate, various names may be used to refer to a component. This specification and claims do not intend to distinguish between components that differ in name but not function. In the following description and in the claims, the terms "include" and "comprise" are used in an open-ended fashion, and thus should be interpreted to mean "include, but not limited to. The description which follows is a preferred embodiment of the disclosure, but is made for the purpose of illustrating the general principles of the disclosure and not for the purpose of limiting the scope of the disclosure. The scope of the present disclosure is to be determined by the terms of the appended claims.
To facilitate an understanding of the embodiments of the present disclosure, the following detailed description is to be considered in conjunction with the accompanying drawings, and the drawings are not to be construed as limiting the embodiments of the present disclosure.
In one embodiment, as shown in fig. 1, the present disclosure provides a method for predicting magnetic particle distribution of a magnetic control software robot, comprising the following steps:
s100: observing the motion state of the magnetic control soft robot to be detected under a specific magnetic field, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state;
s200: and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the distribution of the magnetic particles in the magnetic control software robot.
Compared with the prior art, the embodiment can search the mapping relation between the response and the model structure from the simulation data, so that the traditional mathematical physics calculation method is avoided, and the model structure corresponding to the response can be obtained at the output end only by inputting the expected response at the input end of the network, so that the processes of modeling, parameter sweeping optimization calculation and the like in the traditional design method are omitted, the research and development efficiency is improved, and the overhigh cost caused by blind design is avoided.
In another embodiment, the training process of the prediction model is represented as follows:
s201: under a specific magnetic field, designing magnetic particle distribution in the magnetic control soft robot in different modes, and recording the motion state of the robot corresponding to each design to obtain a first magnetic particle distribution-motion state corresponding relation data set;
in this embodiment, the comsol simulation software is used to simulate the magnetic particle distribution in the magnetically controlled soft robot, so as to obtain the corresponding relationship between the magnetic particle distribution and the motion state of the robot, and the specific simulation process is as follows: first, a helmholtz coil three-dimensional model is built in the comsol as shown in fig. 2, the helmholtz coil can generate a uniform magnetic field at its axial position, and the three-dimensional helmholtz coil can generate a uniform magnetic field of any magnitude and direction at its central position by adjusting the magnitude of current in each coil. In order to accurately obtain the correspondence between the magnetic particle distribution and the motion state of the robot, it is necessary to keep the current in the three-dimensional helmholtz coil stable so that the magnetic field is stable, change the distribution of the magnetic particles at the center of the coil, and as shown in fig. 3(a) to 3(c), the magnetic particles are aggregated in three distribution modes, i.e., a vertical column shown in fig. 3(a), an oblique column shown in fig. 3(b), and a curved column shown in fig. 3(c), while observing and recording the motion state of the robot in each of the above magnetic particle distribution modes.
S202: respectively carrying out digital processing on the distribution and motion state of all the magnetic particles in the first magnetic particle distribution-motion state corresponding relation data set to obtain a second magnetic particle distribution-motion state corresponding relation data set;
in this step, the digital processing of the magnetic particle distribution is performed as follows:
expressing the distribution state of magnetic particles in the robot as a coordinate sequence, for example, when the magnetic particles are in a column-shaped distribution as shown in fig. 3(a), four columns are gridded, and as a result, as shown in fig. 4, each column is divided into 5 uniform parts, wherein the coordinate sequence of the column numbered (i) in fig. 4 is expressed as (5, 5, 2), (5, 5, 4), (5, 5, 6), (5, 5, 8), (5, 5, 10) from low to high; the coordinate sequence of the upright column with the number of (c) is represented as (5, -5, 2), (5, -5, 4), (5, -5, 6), (5, -5, 8), (5, -5, 10) from low to high; the coordinate sequence of the upright column numbered with the third number is represented as (-5, 5, 2), (-5, 5, 4), (-5, 5, 6), (-5, 5, 8), (-5, 5, 10) from low to high; the coordinate sequence of the upright column with the number of (4) is represented as (-5, -5, 2), (-5, -5, 4), (-5, -5, 6), (-5, -5, 8), (-5, -5, 10) from low to high. The coordinates of each column are from the area closest to the origin of the coordinates shown in the figure and from the end of the robot limb to the end, wherein each coordinate in the coordinate sequence represents a cluster of agglomerated magnetic particles in one column.
The motion state is digitally processed in the following way:
under the specific magnetic field of (0, -0.5 pi, 50), the four columns in fig. 3(a) will have some form of bending motion, so that the robot moves forward, and the four columns will finally keep the bending shape as shown in fig. 5(a), and the same way as the digitalization way when the columns are distributed like the columns, the following coordinate sequence is obtained after the curved columns are gridded according to numbers (r) -r: (5, 5, 2), (4.3, 5, 3.8), (2.9, 5, 6.1), (2.4, 5, 7.3), (1.5, 5, 8.5); (-5, 5, 2), (-5.7, 5, 3.8), (-7.1, 5, 6.1), (-7.6, 5, 7.3), (-8.5, 5, 8.5); (-5, -5, 2), (-5.7, -5, 3.8), (-7.1, -5, 6.1), (-7.6, -5, 7.3), (-8.5, -5, 8.5); (5, -5, 2),(4.3, -5,3.8),(2.9, -5,6.1),(2.4, -5,7.3),(1.5, -5, 8.5). Similarly, under a specific magnetic field of (0, 0.5 pi, 50), the robot moves backwards, and the four columns finally keep the reverse bending shape as shown in fig. 5 (b); under a specific magnetic field of (0, 0.5 pi, 110), the robot will perform backward steering motion, and the four columns will eventually maintain a more reversely curved shape as shown in fig. 5 (c).
S203: and dividing the second magnetic particle distribution-motion state corresponding relation data set into a training set and a testing set, training the prediction model through the training set and testing the trained prediction model through the testing set, finishing model training if a testing result meets a preset accuracy rate, and otherwise, optimizing network parameters and training configuration parameters of the prediction model until the training requirements are met.
In this step, 70% of the second magnetic particle distribution-motion state correspondence data set is used as a training set, and the remaining 30% is used as a test set, and the training process is exemplarily illustrated by using a GAN network as a prediction model as shown in fig. 6:
the GAN network comprises a generator and a discriminator, wherein a target motion sequence is used as an input end of the generator, and a predicted distribution coordinate sequence called a pseudo distribution coordinate sequence is obtained through a hidden layer of the generator; the discriminator comprises a trained neural network, a motion sequence is obtained by taking a real distribution sequence in a training set as input, the network can perfectly fit the distribution sequence and the action sequence in a training sample, at the moment, a pseudo distribution coordinate sequence generated in the generator is input into the discriminator, the discriminator can output a pseudo action sequence corresponding to the pseudo distribution coordinate sequence, the pseudo action sequence can generate an error with a target motion sequence, a critical error value is set, when the error between the pseudo action sequence and the target motion sequence is greater than the critical error value, the network can automatically adjust hidden layer parameters in the generator to reduce the error between the pseudo motion sequence generated by the distributed coordinate sequence through the discriminator and the target motion sequence, the process is repeated continuously until the error is less than the critical error value, at this time, the discriminator cannot distinguish whether the pseudo distribution coordinate sequence is a real distribution sequence or a distribution sequence, and at this time, the network training is completed, and the pseudo distribution coordinate sequence generated in the generator is regarded as a real distribution sequence.
It should be noted that after the training of the prediction model is completed, in order to ensure the accuracy of the model prediction, the model still needs to be further verified, and the verification process is as follows: obtaining a distribution mode of certain magnetic particles from the output of a prediction model according to a certain motion state of the robot, constructing the robot according to the distribution mode, then observing whether the actual motion state of the robot is consistent with the motion state input as the model, if not, carrying out coordinate sequence division again, or adjusting network structure and network parameters, such as changing the number of hidden layers and neurons, reselecting an activation function, selecting other types of networks, and the like.
In another embodiment, the present disclosure further provides a method for predicting magnetic field distribution of a magnetic control software robot, including the following steps:
s1000: observing the motion state of the magnetic control soft robot to be detected under the distribution of specific magnetic particles, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state;
s2000: and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the magnetic field of the magnetic control software robot.
In another embodiment, the training process of the prediction model is represented as follows:
s2001: under the distribution of specific magnetic particles, changing the size and direction of a magnetic field outside the magnetically controlled soft robot, and recording the motion state of the robot corresponding to the magnetic field, so as to obtain a first magnetic field distribution-motion state corresponding relation data set;
as with the above-mentioned correspondence between magnetic particle distribution and motion state, the embodiment also simulates the magnetic field distribution outside the magnetically controlled soft robot through the comsol simulation software, and further obtains the correspondence between the magnetic field distribution and the motion state of the robot, and the specific simulation process is as follows: first, a Helmholtz coil three-dimensional model as shown in FIG. 2 is built in comsol, a magnetic particle distribution mode is selected, for example, the magnetic particles as shown in FIG. 3(a) are distributed in a column shape, under the condition of the magnetic particle distribution, the current in the Helmholtz coil is changed at different moments t, and the current is recorded at each moment t i And (b) lowering the motion state of the robot containing the magnetic particles, thereby obtaining the correspondence between the magnetic field distribution and the motion state when the magnetic particles are distributed in the shape of a pillar as shown in fig. 5(a), that is, when the magnetic field distribution is (0, -0.5 pi, 50), the robot moves forward.
S2002: respectively carrying out digital processing on all the magnetic field distribution and the motion state in the first magnetic field distribution-motion state corresponding relation data set to obtain a second magnetic field distribution-motion state corresponding relation data set;
in this step, the digital processing of the magnetic field distribution is performed as follows:
the magnetic field as a driving source is kept unchanged in the driving process, and the magnetic field comprises two states of magnitude and direction. If the uniform magnetic field at the center of the three-dimensional Helmholtz coil is expressed as (0.25 π, 0.25 π, 100), the direction is expressed in terms of two angles, the first of which represents the angle (ranging from 0 to 2 π) at which the magnetic field direction projects in the xoy plane; the second numerical table is the angle (range-pi to pi) of rotation from the positive half axis to the negative half axis from the z-axis direction with the origin of the magnetic field direction as the center; the third value represents the magnetic field strength (in Gs).
The motion state is digitally processed in the following way:
in this embodiment, the motion state is digitized in the same manner as the motion state in the magnetic particle distribution manner prediction step, and as shown in fig. 5(a), four columns may have a certain form of bending motion, and the four columns may finally keep a bending shape, and the four columns are digitized in the same manner as the four columns before the motion, and are subjected to gridding to obtain a coordinate sequence having the same number as the coordinates before the motion, where the numbers from (i) to (iv) are respectively: (5, 5, 2), (4.3, 5, 3.8), (2.9, 5, 6.1), (2.4, 5, 7.3), (1.5, 5, 8.5); (-5, 5, 2), (-5.7, 5, 3.8), (-7.1, 5, 6.1), (-7.6, 5, 7.3), (-8.5, 5, 8.5); (-5, -5, 2), (-5.7, -5, 3.8), (-7.1, -5, 6.1), (-7.6, -5, 7.3), (-8.5, -5, 8.5); (5, -5, 2), (4.3, -5, 3.8), (2.9, -5, 6.1), (2.4, -5, 7.3), (1.5, -5, 8.5); at this time, the robot moves forward, and the corresponding magnetic field is (0, -0.5 π, 50). Under different magnetic fields, the robot can generate different motion states, such as forward, backward or steering, for example, the column numbered as (r) in fig. 3(a) generates different bending under different magnetic fields, and the corresponding relationship between the magnetic field distribution and the motion states is shown in table 1:
TABLE 1
S2003: and dividing the second magnetic field distribution-motion state corresponding relation data set into a training set and a testing set, training the prediction model through the training set, testing the trained prediction model through the testing set, finishing the model training if the test result meets the preset accuracy, and otherwise, optimizing the network parameters and the training configuration parameters of the prediction model until the training requirements are met.
In this step, 70% of the second magnetic field distribution-motion state correspondence data set is used as a training set, and the remaining 30% is used as a test set, and a DNN network shown in fig. 7 is used as a prediction model to exemplarily explain the training process:
1. initializing model parameters, namely the weight of each layer, establishing a loss function, and taking a divided training set as the input of a prediction model;
2. calculating to obtain a pseudo magnetic field based on the initial weight and the cross-network structure, comparing the pseudo magnetic field with the real magnetic field in the training set, namely, subtracting through a loss function to obtain a loss value, and meanwhile, establishing a loss critical value, wherein the loss value must be smaller than the loss critical value, if the loss value is larger than the loss critical value, the weight in the model needs to be adjusted through back propagation, and when the loss generated by all samples in the training set is smaller than the loss critical value, the model is trained completely, and at this time, the model has better fitting capability. At this time, a completely new coordinate sequence representing the motion state is input into the model, so that a completely new magnetic field distribution can be obtained, for example, the coordinate sequences (5, 5, 2), (5.7, 5, 3.8), (7.1, 5, 6.1), (7.6, 5, 7.3), (8.5, 5, 8.5) in a certain state of the robot are represented; (-5, 5, 2), (-4.3, 5, 3.8), (-2.9, 5, 6.1), (-2.4, 5, 7.3), (-1.5, 5, 8.5); (-5, -5, 2), (-4.3, -5, 3.8), (-2.9, -5, 6.1), (-2.4, -5, 7.3), (-1.5, -5, 8.5); (5, -5, 2), (5.7, -5, 3.8), (7.1, -5, 6.1), (7.6, -5, 7.3), (8.5, -5, 8.5) the magnetic field distribution available for input into the model was (0, 0.5 pi, 50), and it can be seen that the model enables prediction of the magnetic field distribution.
It should be noted that after the training of the prediction model is completed, in order to ensure the accuracy of the model prediction, the model still needs to be further verified, and the verification process is as follows: obtaining a certain magnetic field distribution mode from the output of the prediction model according to a certain motion state of the robot, constructing a magnetic field environment where the robot moves according to the distribution mode, then observing whether the actual motion state of the robot is consistent with the motion state input as the model, if not, carrying out coordinate sequence division again, or adjusting network structure and network parameters, such as changing the number of hidden layers and neurons, activating functions, reselecting other types of networks, and the like.
The present disclosure has been described in detail, and the principles and embodiments of the present disclosure have been explained herein by using specific examples, which are provided only for the purpose of helping understanding the method and the core concept of the present disclosure; meanwhile, for those skilled in the art, according to the idea of the present disclosure, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present description should not be construed as a limitation to the present disclosure.
Claims (8)
1. A magnetic control software robot magnetic particle distribution prediction method comprises the following steps:
s100: observing the motion state of the magnetic control soft robot to be detected under a specific magnetic field, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state;
s200: and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the distribution of the magnetic particles in the magnetic control software robot.
2. The method of claim 1, wherein preferably the predictive model includes, and is not limited to, a DNN, RNN or GAN network.
3. The method of claim 1, wherein the training process of the predictive model comprises the steps of:
s201: under a specific magnetic field, designing magnetic particle distribution in the magnetic control soft robot in different modes, and recording the motion state of the robot corresponding to each design to obtain a first magnetic particle distribution-motion state corresponding relation data set;
s202: respectively carrying out digital processing on the distribution and motion states of all the magnetic particles in the first magnetic particle distribution-motion state corresponding relation data set to obtain a second magnetic particle distribution-motion state corresponding relation data set;
s203: and dividing the second magnetic particle distribution-motion state corresponding relation data set into a training set and a testing set, training the prediction model through the training set and testing the trained prediction model through the testing set, finishing model training if a testing result meets a preset accuracy rate, and otherwise, optimizing network parameters and training configuration parameters of the prediction model until the training requirements are met.
4. The method according to claim 3, wherein the step S202 of digitizing the magnetic particle distribution comprises: and expressing the part containing the magnetic particles by using a coordinate sequence, and meshing the part containing the magnetic particles to obtain coordinates of each node, wherein each coordinate expresses a cluster of agglomerated magnetic particles.
5. The method according to claim 3, wherein the step S202, the digitizing the motion state comprises: and expressing the magnetic particles corresponding to the final motion state by using a coordinate sequence, and performing grid division on the magnetic particles to obtain coordinates of each node, wherein each coordinate expresses a cluster of agglomerated magnetic particles.
6. A magnetic control software robot magnetic field distribution prediction method comprises the following steps:
s1000: observing the motion state of the magnetic control soft robot to be detected under the distribution of specific magnetic particles, and carrying out digital processing on the motion state to obtain a state quantity corresponding to the motion state;
s2000: and inputting the state quantity serving as an input variable into a trained prediction model to obtain an output variable representing the magnetic field of the magnetic control software robot.
7. The method of claim 6, wherein the predictive model includes, and is not limited to, a DNN, RNN, or GAN network.
8. The method of claim 6, wherein the training process of the predictive model is represented as follows:
s2001: under the distribution of specific magnetic particles, changing the size and direction of a magnetic field outside the magnetically controlled soft robot, and recording the motion state of the robot corresponding to the magnetic field, so as to obtain a first magnetic field distribution-motion state corresponding relation data set;
s2002: respectively carrying out digital processing on all the magnetic field distributions and the motion states in the first magnetic field distribution-motion state corresponding relation data set to obtain a second magnetic field distribution-motion state corresponding relation data set;
s2003: and dividing the second magnetic field distribution-motion state corresponding relation data set into a training set and a testing set, training the prediction model through the training set, testing the trained prediction model through the testing set, finishing the model training if the test result meets the preset accuracy, and otherwise, optimizing the network parameters and the training configuration parameters of the prediction model until the training requirements are met.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111553563.XA CN114970305B (en) | 2021-12-17 | 2021-12-17 | Magnetic particle distribution prediction method for magnetically controlled soft robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111553563.XA CN114970305B (en) | 2021-12-17 | 2021-12-17 | Magnetic particle distribution prediction method for magnetically controlled soft robot |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114970305A true CN114970305A (en) | 2022-08-30 |
| CN114970305B CN114970305B (en) | 2024-01-19 |
Family
ID=82974841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111553563.XA Active CN114970305B (en) | 2021-12-17 | 2021-12-17 | Magnetic particle distribution prediction method for magnetically controlled soft robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114970305B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117233520A (en) * | 2023-11-16 | 2023-12-15 | 青岛澎湃海洋探索技术有限公司 | AUV propulsion system fault detection and evaluation method based on improved Sim-GAN |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2004063831A2 (en) * | 2003-01-15 | 2004-07-29 | Bracco Imaging S.P.A. | System and method for optimization of a database for the training and testing of prediction algorithms |
| CN108621159A (en) * | 2018-04-28 | 2018-10-09 | 首都师范大学 | A kind of Dynamic Modeling in Robotics method based on deep learning |
| CN113246168A (en) * | 2021-04-20 | 2021-08-13 | 北京理工大学 | Magnetic drive deformation composite structure, preparation method, soft robot and application |
-
2021
- 2021-12-17 CN CN202111553563.XA patent/CN114970305B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2004063831A2 (en) * | 2003-01-15 | 2004-07-29 | Bracco Imaging S.P.A. | System and method for optimization of a database for the training and testing of prediction algorithms |
| CN108621159A (en) * | 2018-04-28 | 2018-10-09 | 首都师范大学 | A kind of Dynamic Modeling in Robotics method based on deep learning |
| CN113246168A (en) * | 2021-04-20 | 2021-08-13 | 北京理工大学 | Magnetic drive deformation composite structure, preparation method, soft robot and application |
Non-Patent Citations (2)
| Title |
|---|
| 姚为;廉城;程立;: "回声状态网络动态预测模型的分叉迭代预测方法", 小型微型计算机系统, no. 10 * |
| 王馨;魏树峰;陈昌友;徐建省;张玉霞;宋涛;: "用于磁性微机器人的外磁场调控系统设计与研制", 电工电能新技术, no. 02 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117233520A (en) * | 2023-11-16 | 2023-12-15 | 青岛澎湃海洋探索技术有限公司 | AUV propulsion system fault detection and evaluation method based on improved Sim-GAN |
| CN117233520B (en) * | 2023-11-16 | 2024-01-26 | 青岛澎湃海洋探索技术有限公司 | AUV propulsion system fault detection and evaluation method based on improved Sim-GAN |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114970305B (en) | 2024-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112308961B (en) | Robot rapid robust three-dimensional reconstruction method based on layered Gaussian mixture model | |
| CN113031528B (en) | Multi-legged robot non-structural ground motion control method based on depth certainty strategy gradient | |
| CN110543727B (en) | Improved particle swarm algorithm-based omnidirectional mobile intelligent wheelchair robot parameter identification method | |
| CN110488611B (en) | Bionic robot fish motion control method, controller and bionic robot fish | |
| US7831418B1 (en) | Autonomous experimental design optimization | |
| CN112036062A (en) | Metal material bending forming rebound angle prediction method | |
| CN114970305A (en) | Prediction method for magnetic particle distribution of magnetic control software robot | |
| CN114722731B (en) | Method and system for predicting drag viscous drag resistance of towed body | |
| CN117369244A (en) | Welding gun position control optimization method based on welding robot | |
| Serrano-Pérez et al. | Offline robust tuning of the motion control for omnidirectional mobile robots | |
| CN115248591A (en) | UUV path planning method based on hybrid initialization Hui wolf particle swarm algorithm | |
| Li et al. | Dynamics and workspace analysis of a multi-robot collaborative towing system with floating base | |
| CN113722853B (en) | Group intelligent evolutionary engineering design constraint optimization method for intelligent computation | |
| CN104038135B (en) | A kind of torque motor structure parameter optimization method | |
| CN115758766A (en) | Continuous-discontinuous numerical model connecting method | |
| CN114967472A (en) | Unmanned aerial vehicle trajectory tracking state compensation depth certainty strategy gradient control method | |
| Roslan et al. | Sine-Based Barnacle Mating Optimization for PD Control of an Inverted Pendulum System | |
| CN114491790A (en) | MAML-based pneumatic modeling method and system | |
| Lin et al. | The future of environmental performance architectural design based on human-computer interaction | |
| CN117933104B (en) | Solid attitude and orbit control engine gas regulating valve pressure correction method | |
| CN114186442B (en) | Lattice material parameter optimization method based on neural network model and numerical simulation | |
| CN115906659B (en) | Neutron transport method based on neural network for nuclear reactor critical design | |
| CN115146389B (en) | Permanent magnet magnetic levitation train dynamics feature modeling method | |
| CN115130338B (en) | Magnetic suspension workbench magnetic force model construction method suitable for parallel computing | |
| CN114734446B (en) | Manipulator high-precision position control method based on improved reinforcement learning algorithm |
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 |