CN117325186A - Sampling mechanical arm path planning-based method and system - Google Patents
Sampling mechanical arm path planning-based method and system Download PDFInfo
- Publication number
- CN117325186A CN117325186A CN202311620923.2A CN202311620923A CN117325186A CN 117325186 A CN117325186 A CN 117325186A CN 202311620923 A CN202311620923 A CN 202311620923A CN 117325186 A CN117325186 A CN 117325186A
- Authority
- CN
- China
- Prior art keywords
- sampling
- vector
- path
- model
- module
- 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
- 238000005070 sampling Methods 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 46
- 239000013598 vector Substances 0.000 claims abstract description 101
- 238000009826 distribution Methods 0.000 claims abstract description 41
- 238000013139 quantization Methods 0.000 claims abstract description 27
- 230000006870 function Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000004519 manufacturing process Methods 0.000 abstract description 7
- 230000004931 aggregating effect Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 3
- 230000004075 alteration Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
-
- 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
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/163—Programme controls characterised by the control loop learning, adaptive, model based, rule based expert control
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a method and a system for path planning based on a sampling mechanical arm, comprising the steps of obtaining a discretization vector set of a planning space by taking a transducer model as an encoderCollecting the discretized vectorEach value in the model is obtained through a multi-layer perceptron model decoder to obtain Gaussian distribution corresponding to each value, and a vector quantization module is generated; aggregating discretized vectors in the vector quantization moduleThe Gaussian distribution is combined with the starting position, the target position and the surrounding environment of the given mechanical arm to generate a sampling area, and a sampling area generation module is generated; based on a sampling area generating module, a mechanical arm movement path is obtained according to the sampling area, and a path is generatedAnd a path planning module. Therefore, the optimal path can be generated more quickly, the problems of generalization and expansibility in the path planning of the mechanical arm are solved, and the production efficiency of a factory can be effectively improved.
Description
Technical Field
The invention relates to the technical field of mechanical arms, in particular to a method and a system for path planning based on a sampling mechanical arm.
Background
With the introduction of intelligent production lines, more and more tasks in new energy factories are completed by robots, but due to the huge volume and complex structure of industrial robots, workers working nearby may face significant risks, and thus related robot manufacturers have designed compliant mechanical arms to strictly follow the production flow to prevent safety production problems. The robot arm may collide with an external object or a worker inevitably during movement, and thus, related manufacturers have designed a series of path planning algorithms to avoid obstacles.
To achieve efficient sampling, previous research efforts have reduced search space by manually designed heuristics or parameterized functions, thereby shortening planning time. The most advanced planning methods today employ searching paths in the elliptical region between the starting point and the target point, but for high dimensional space, sampling using these heuristics still results in many samples not being used to construct the trajectory. Machine learning based methods utilize data in existing planning data sets to optimize planning in similar environments, such as researchers using Gaussian Mixture Models (GMMs) to learn the distribution of previous path plans, but do not have good scalability in complex environments.
The traditional path planning algorithm lacks expandability in a high-dimensional space, and the existing path planning method of the sampling mechanical arm lacks generalization in a training scene, so that the mechanical arm path planning method for different environments in the prior art is low in generalization and expansibility.
Disclosure of Invention
The embodiment of the application solves the problems of generalization and expansibility of the mechanical arm path planning method for different environments in the prior art, realizes faster generation of an optimal path and solves the problem of generalization and expansibility in mechanical arm path planning by providing the method and the system based on sampling mechanical arm path planning.
In a first aspect, embodiments of the present application provide a method for sampling-based path planning for a robotic arm, including,
s1, obtaining a discretized vector set of a planning space by using a transducer model as an encoderThe discretization vector set +.>Each value in the model is obtained through a multi-layer perceptron model decoder to obtain Gaussian distribution corresponding to each value, and a vector quantization module is generated;
s2, collecting discrete vectors in the vector quantization moduleThe Gaussian distribution is combined with the starting position, the target position and the surrounding environment of the given mechanical arm to generate a sampling area, and a sampling area generation module is generated;
s3, based on a sampling area generating module, obtaining a mechanical arm movement path according to the sampling area, generating a path planning module, and giving out the sampling area by improving an RRT algorithmAnd generating a corresponding mechanical arm path.
Further, in step S1, further comprising,
s11, inputting a group of mechanical tracks through an encoder, wherein the encoder linearly projects each input state in the tracks to a potential space through learning a mode in a mechanical track sequence to obtain a group of potential vectors;
s12, setting a dictionary set, discretizing each potential vector to obtain a corresponding dictionary value, finding the nearest vector in the dictionary set, and replacing the vector with the dictionary value to obtain a group of vector sets;
and S13, adding the group of vector sets into a start vector and an end vector to obtain a final group of vector sets, and mapping the final group of vector sets into a group of Gaussian distribution by using a multi-layer perceptron model decoder.
Further, each dictionary value output by the encoder is decoded by a multi-layer perceptron model decoderMapping into a parameterized distribution +.>The parameterized distribution is selected as Gaussian distribution, and the multi-layer perceptron model decoder outputs a mean value and covariance matrix of the Gaussian distribution>,
Based on each dictionary value output by the encoderSaid dictionary value->The corresponding mean and covariance matrix is +.>,/>。
Further, the result output by the multi-layer perceptron model decoder of the penultimate multi-layer perceptron model is obtained through a single linear layerAnd->The method comprises the steps of carrying out a first treatment on the surface of the Diagonal matrix->After passing through the linear layer, it is processed through a soft-plus function.
Further, in step S2, the sampling region generation module further includes,
s21, the starting point and the target point of the mechanical arm are paired through a cross-attention transducer modelEnvironment->Representation ofPotential vector->;
S22, predicting the sampling region through an autoregressive transducer model to obtain a dictionary set of the sampling region corresponding to the planning spaceIndex value in (a)Each index valueGenerating a distribution by the autoregressive model; the following formula is shown:
;
wherein p is i Is h j Based on the corresponding probabilities obtained by the autoregressive transformers,is the vector corresponding to the dictionary set, M is the potential vector obtained according to the target point pair and the sampling area, < >>Representing an autoregressive transducer model, N is the dictionary set size.
Further, the method comprises the steps of,
training the cross-attention and autoregressive-based Transformer models using cross entropy loss;
;
wherein,is a Croneck function,/->Is the output of the autoregressive model, +.>Is the true index +.>Corresponding potential dictionary vectors; dictionary values for each prediction are mapped in a planning space using a learned decoding from a vector quantization moduleConversion to the corresponding Gaussian distribution>,/>Representing the number of index values in the sample generation region generation module.
Further, the sampling region generation module generates an index sequence according to the cross-attention transducer model, the auto-regression-based transducer model and the planning space dictionary setBecause each index takes N values in the planning space dictionary set, an optimal index sequence is obtained, as shown in the following formula:
;
wherein the method comprises the steps ofIs the target index,/-, is>Representing an autoregressive transducer model.
Further, in step S3, further comprising,
based on the resulting index sequenceAfter that, useDecoder model of vector quantization module to correspond potential dictionary vector according to index value>Generating a set of sampling regions->Wherein we calculate using the mixture Gaussian model to get the sampling region +.>The following formula is shown:
。
further, in the path planning module, the method further comprises,
improving RRT algorithm, and inputting a starting position q by the path planning module s Target position q g Sampling regionNumber of cycles k; first the start position +.>Add sample set +.>In each cycle, from the sampling area +.>Generating a random sampling node->According to->Sample set +.>Find the nearest point +.>The method comprises the steps of carrying out a first treatment on the surface of the If->And->If an active path exists, a link is established and +.>Join set->Ending the current cycle;
otherwise, randomly obtaining a number, if the number is larger than the threshold value b, according to the target position q g And sample setFind nearest point +.>If->And->If an active path exists, a link is established and +.>Join set->And jump out of the loop, return the set +.>The method comprises the steps of carrying out a first treatment on the surface of the If no valid path exists, the loop continues.
In another aspect, a system for sampling-based path planning for a robotic arm includes,
vector quantization module for obtaining discretized vector set of planning space by using transducer model as encoderThe discretization vector set +.>Each value in the model is obtained through a multi-layer perceptron model decoder to obtain Gaussian distribution corresponding to each value, and a vector quantization module is generated;
a sampling region generation module for collecting the discretized vector set in the vector quantization moduleThe Gaussian distribution is combined with the starting position, the target position and the surrounding environment of the given mechanical arm to generate a sampling area, and a sampling area generation module is generated;
and the path planning module is used for obtaining a mechanical arm movement path according to the sampling area and generating the path planning module based on the sampling area generation module.
By improving RRT algorithm, the sampling area is givenAnd generating a corresponding mechanical arm path.
The beneficial effects are that:
1. discretizing a planning space through a vector quantization module, representing the discretized planning space as a learnable vector, obtaining a sampling area with optimal given planning problems through autoregressive model learning, and finally obtaining a final path through an improved RRT algorithm.
2. The large planning space is divided into discrete vector sets, an optimal sampling area is generated according to the environment, the initial position and the target position, and the path is generated by matching with an improved rapid search random tree algorithm, so that the optimal path can be generated more rapidly, the problems of generalization and expansibility in the path planning of the mechanical arm are solved, and the production efficiency of a factory can be effectively improved.
Drawings
FIG. 1 is a flow chart of a method for path planning based on a sampling robot;
FIG. 2 is a diagram of a method architecture based on sampling robot path planning;
fig. 3 is a diagram of a vector quantization module and a sampling region generation module.
Detailed Description
The embodiment of the application solves the problems of generalization and expansibility of the mechanical arm path planning method for different environments in the prior art, realizes faster generation of an optimal path and solves the problem of generalization and expansibility in mechanical arm path planning by providing the method and the system based on sampling mechanical arm path planning.
In order to solve the problem of crosstalk, the technical solution in the embodiment of the present application is as follows:
the device comprises a vector quantization module, a sampling area generation module and a path planning module. The vector quantization module obtains discrete vectors of the planning space by using a transducer model by using the thought of VQ-VAE, and obtains Gaussian distribution of each discrete vector by using a multi-layer perceptron model. And the sampling area generating module is used for learning environmental information by using the cross attention model, obtaining sequences with maximum probability by using the autoregressive model, and obtaining the corresponding distribution of each sequence by using the decoder so as to generate an optimal sampling area for the path planning module. And the path planning module is used for generating an optimal path for the mechanical arm by improving the RTT algorithm and using the sampling area given by the sampling area generating module.
In order to better understand the above technical solutions, the following detailed description will refer to the accompanying drawings and specific embodiments.
Referring to fig. 1 and 2, a method for path planning based on a sampling mechanical arm includes
S1, obtaining a discretized vector set of a planning space by using a transducer model as an encoderThe discretization vector set +.>Each value in the model is obtained through a multi-layer perceptron model decoder to obtain gauss corresponding to each valueDistributing to generate a vector quantization module;
in particular, the vector quantization module comprises,
s11, inputting a group of mechanical tracks through an encoder, wherein the encoder linearly projects each input state in the tracks to a potential space through learning a mode in a mechanical track sequence to obtain a group of potential vectors;
s12, a dictionary set is set, each potential vector is discretized to obtain a corresponding dictionary value, the closest vector is found in the dictionary set, and the vector and the dictionary value are replaced to obtain a group of vector sets;
s13, adding the set of vectors into a start vector and an end vector to obtain a final set of vectors, and mapping the final set of vectors into a set of Gaussian distributions P by using a multi-layer perceptron model decoder.
Specifically, the vector quantization module is mainly used for vectorizing the whole planning space by learning the data set distribution of the existing mechanical arm track.
Specifically, as shown in FIG. 3, the present invention uses a transducer model as an encoder with a set of robot arm trajectories as input, taking the concept of VQ-VAE into considerationThe encoder model converts each state in the track into a valid representation, i.e. a set of potential vectors +.>。
For each input stateThe encoder models all project it linearly to the potential space +.>Obtain->. Setting a dictionary set->Which comprises->Variable(s)>Representing the entire planning space. Vector +.>We need to discretize it to get the corresponding dictionary value. According to formula 1, find the vector nearest thereto in the dictionary set +.>Substitution is performed.
;
Wherein the method comprises the steps ofIs in combination with->The most relevant vector.
Adding a start and end vectorAnd->. For a set of input trajectories of the robot arm, the output of the encoder is finally a set of vectors set +.>。
Using a multi-layer perceptron (MLP) model as a decoder, each of the encoder outputsMapping into a parameterized distribution +.>The invention selects the parameterized distribution as Gaussian distribution, and the decoder outputs the mean value and covariance matrix of the Gaussian distribution +.>. Because of->Is the dictionary value, and the corresponding mean and covariance matrix is,/>For simplicity hereafter use +.>And->And (3) representing. To ensure that the covariance matrix remains positive during training, equation 2 is used to decompose it into a product of a lower triangular matrix and its transpose,
;
wherein the method comprises the steps ofIs a lower triangular matrix, ">Is a diagonal matrix. The present invention obtains +_by passing the result of the decoder model penultimate layer MLP (Multi-layer perceptron) output through a separate linear layer>And->For->After the linear layer, it is also processed by a soft-plus function to ensure that its value is positive. The soft-plus function is a smooth nonlinear function that is similar to a ReLU (modified linear unit) but has a continuous gradient at near zero values, which helps to alleviate the problem of gradient extinction or gradient explosion during training.
The reconstruction loss of the present invention is shown in equation 3:
;
wherein,is constant, & lt>It is desirable that the second term of the formula is to prevent the distribution from overfitting each data batch, as smaller batches do not cover the entire planning space.
Training is carried out by feeding different mechanical arm tracks, and finally a discretization vector set of the whole planning space can be obtainedAnd the distribution corresponding to each value in the set provides corresponding data for the inquiry of the subsequent sampling area generation module.
S2, collecting discrete vectors in the vector quantization moduleThe Gaussian distribution is combined with the starting position, the target position and the surrounding environment of the given mechanical arm to generate a sampling area, and a sampling area generation module is generated;
specifically, the sampling region generating module generates a sampable region from the starting position to the target position for the starting position, the target position, the surrounding environment and the distribution corresponding to each value in the planning space dictionary set and the dictionary set obtained by the vector quantization module. It comprises two models, a Cross-Attention (Cross-Attention) based transducer model and an Autoregressive (Autoregressive) based transducer model.
Wherein the cross-attention based transducer model is used to pair the starting position and the target position of the mechanical armEnvironment->Expressed as potential vector +.>. Specifically, it first generates the coding of the surrounding environment by extracting features from the point cloud +.>Wherein->For start and target states, linear projection is used to obtainIn the cross-attention model, +.>As key variable, ++>As query variable to finally generate the potential vector +.>。
Autoregressive-based transducer modelFor predicting sampling areas, obtaining sampling areas corresponding to the dictionary set in planning space>Index value +.>. For each index value +.>The autoregressive model outputs a distribution as shown in equation 4:
;
wherein p is i Hj is the corresponding probability obtained according to the autoregressive Transform, z is the vector corresponding to the dictionary set, M is the potential vector obtained according to the target point pair and the environment, and the pie represents the autoregressive Transform model. All pi add up to 1, i.e. the sum of probabilities is 1, N is the dictionary set size.
The two models of the present module will be trained using cross entropy loss as shown in equation 5:
;
wherein,is a Croneck function,/->Is the output of the autoregressive model, +.>Is the true index +.>Corresponding potential dictionary vectors. Using the learned decoding from the vector quantization module, we can +_dictionary value for each prediction in the planning space>Conversion into a pairGauss distribution of response->,/>Representing the number of index values in the sample generation region generation module.
The training process of the sampling area generation module is that for a new path planning problem, such as the prediction process shown in fig. 1, the module generates an index sequence according to the two models and the planning space dictionary setSince each index can take +.>The value, therefore we need to add the sequence +.>The probability is maximized to obtain the optimal index sequence as shown in equation 6.
(6);
Wherein the method comprises the steps ofIs the target index,/-, is>Is the probability obtained by equation 4.
In the index sequence where the highest probability is obtainedAfter that, we use the decoder model of the vector quantization module to base the potential vector +.>Generating a set of distribution sets->Wherein we calculate the distribution +.>As shown in equation 7:
;
so that each index value has its corresponding sampling area. From the maximum probability sequence we can get a set of sampling regions from the starting position to the target position in the planning space.
And S3, based on the sampling area generating module, obtaining a mechanical arm movement path according to the sampling area, and generating a path planning module.
Specifically, a path planning module improves the RRT algorithm, which inputs the starting position q s Target position q g Sampling regionNumber of cycles k; first the start position +.>Add sample set +.>In each cycle, from the sampling area +.>Generating a random sampling node->According to->Sample set +.>Find the nearest point +.>The method comprises the steps of carrying out a first treatment on the surface of the If->And (3) withIf an active path exists, a link is established and +.>Join set->Ending the current cycle;
otherwise, randomly obtaining a number, if the number is larger than the threshold value b, according to the target position q g And sample setFind nearest point +.>If->And->If an active path exists, a link is established and +.>Join set->And jump out of the loop, return the set +.>The method comprises the steps of carrying out a first treatment on the surface of the If no valid path exists, the loop continues.
Unlike the conventional RRT algorithm, the present scheme does not simply extend the current node by a small distance, but checks whether a valid path exists between the current node and the sampling node; if a valid path exists, a link is established. The method can generate the optimal path more quickly, solve the problems of generalization and expansibility in the path planning of the mechanical arm, and effectively improve the production efficiency of the factory.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the invention
Clear spirit and scope. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (10)
1. A method for planning a path based on a sampling mechanical arm is characterized by comprising the following steps of,
s1, obtaining a discretized vector set of a planning space by using a transducer model as an encoderThe discretization vector set +.>Each value in the model is obtained through a multi-layer perceptron model decoder to obtain Gaussian distribution corresponding to each value, and a vector quantization module is generated;
s2, collecting discrete vectors in the vector quantization moduleThe Gaussian distribution is combined with the starting position, the target position and the surrounding environment of the given mechanical arm to generate a sampling area, and a sampling area generation module is generated;
s3, based on a sampling area generating module, obtaining a mechanical arm movement path according to the sampling areaThe path is generated, a path planning module is generated, and a sampling area is given by improving the RRT algorithmAnd generating a corresponding mechanical arm path.
2. The method for path planning based on a sampling robot according to claim 1, further comprising, in step S1,
s11, inputting a group of mechanical tracks through an encoder, wherein the encoder linearly projects each input state in the tracks to a potential space through learning a mode in a mechanical track sequence to obtain a group of potential vectors;
s12, setting a dictionary set, discretizing each potential vector to obtain a corresponding dictionary value, finding the nearest vector in the dictionary set, and replacing the vector with the dictionary value to obtain a group of vector sets;
and S13, adding the group of vector sets into a start vector and an end vector to obtain a final group of vector sets, and mapping the final group of vector sets into a group of Gaussian distribution by using a multi-layer perceptron model decoder.
3. The method for sampling-based path planning of claim 2, further comprising,
each dictionary value output by the encoder is decoded by a multi-layer perceptron model decoderMapping into a parameterized distribution +.>The parameterized distribution is selected as Gaussian distribution, and the multi-layer perceptron model decoder outputs a mean value and covariance matrix of the Gaussian distribution>,
Based on each dictionary value output by the encoderSaid dictionary value->The corresponding mean and covariance matrices are,/>。
4. A method for sampling-based path planning as defined in claim 3, further comprising,
the result output by the multi-layer perceptron model of the penultimate layer of the multi-layer perceptron model decoder is obtained through a single linear layerAnd->The method comprises the steps of carrying out a first treatment on the surface of the Diagonal matrix->After passing through the linear layer, it is processed through a soft-plus function.
5. The method for path planning based on a sampling robot according to claim 1, further comprising, in step S2,
s21, the starting point and the target point of the mechanical arm are paired through a cross-attention transducer modelEnvironment->Expressed as potential vector +.>;
S22, predicting the sampling region through an autoregressive transducer model to obtain a dictionary set of the sampling region corresponding to the planning spaceIndex value in (a)Each index valueGenerating a distribution by the autoregressive model; the following formula is shown:
;
wherein p is i Is h j Based on the corresponding probabilities obtained by the autoregressive transformers,is the vector corresponding to the dictionary set, M is the potential vector obtained according to the target point pair and the sampling area, < >>Representing an autoregressive transducer model, N is the dictionary set size.
6. The method for sampling-based path planning of claim 5, comprising,
training the cross-attention and autoregressive-based Transformer models using cross entropy loss;
;
wherein,is a kronecker function, which,is the output of the autoregressive model,is a true indexCorresponding potential dictionary vectors; converting each predicted dictionary value into a corresponding gaussian distribution in a planning space using a learned decoding from a vector quantization module,Representing the number of index values in the sample generation region generation module.
7. The method for sampling-based path planning of claim 6, comprising,
the sampling area generation module generates an index sequence according to the cross-attention transducer model and the auto-regression-based transducer model and a planning space dictionary setBecause each index takes N values in the planning space dictionary set, an optimal index sequence is obtained, as shown in the following formula:
;
wherein the method comprises the steps ofIs the target index,/-, is>Representing an autoregressive transducer model.
8. The method of claim 7, further comprising, in step S3,
based on the resulting index sequenceThen, a decoder model of the vector quantization module is used to generate potential dictionary vectors corresponding to the index values +.>Generating a set of sampling regions->Wherein the sampling area +.>The following formula is shown:
。
9. the method of claim 1, wherein, in the path planning module, the modified RRT algorithm comprises,
improving RRT algorithm, and inputting a starting position q by the path planning module s Target position q g Sampling regionNumber of cycles k; will firstStart position->Add sample set +.>In each cycle, from the sampling area +.>Generating a random sampling nodeAccording to->Sample set +.>Find the nearest point +.>The method comprises the steps of carrying out a first treatment on the surface of the If->And->If an active path exists, a link is established and +.>Join set->Ending the current cycle;
otherwise, randomly obtaining a number, if the number is larger than the threshold value b, according to the target position q g And sample setFind nearest point +.>If->And->If an active path exists, a link is established and +.>Join set->And jump out of the loop, return the set +.>The method comprises the steps of carrying out a first treatment on the surface of the If no valid path exists, the loop continues.
10. A system based on sampling mechanical arm path planning is characterized by comprising,
vector quantization module for obtaining discretized vector set of planning space by using transducer model as encoderThe discretization vector set +.>Each value in the model is obtained through a multi-layer perceptron model decoder to obtain Gaussian distribution corresponding to each value, and a vector quantization module is generated;
a sampling region generation module for collecting the discretized vector set in the vector quantization moduleThe Gaussian distribution is combined with the starting position, the target position and the surrounding environment of the given mechanical arm to generate a sampling area, and a sampling area generation module is generated;
the path planning module is used for obtaining a mechanical arm movement path according to the sampling area and generating a path planning module based on the sampling area generation module;
by improving RRT algorithm, the sampling area is givenAnd generating a corresponding mechanical arm path.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311620923.2A CN117325186A (en) | 2023-11-30 | 2023-11-30 | Sampling mechanical arm path planning-based method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311620923.2A CN117325186A (en) | 2023-11-30 | 2023-11-30 | Sampling mechanical arm path planning-based method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117325186A true CN117325186A (en) | 2024-01-02 |
Family
ID=89283409
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311620923.2A Pending CN117325186A (en) | 2023-11-30 | 2023-11-30 | Sampling mechanical arm path planning-based method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117325186A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118143950A (en) * | 2024-04-16 | 2024-06-07 | 江苏宁昆机器人智能科技有限公司 | Industrial robot accurate operation method and system based on advanced perception |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2634626A1 (en) * | 2008-07-09 | 2010-01-09 | Gerard Voon | Neural links and artificial intelligence |
CN106780639A (en) * | 2017-01-20 | 2017-05-31 | 中国海洋大学 | Hash coding method based on the sparse insertion of significant characteristics and extreme learning machine |
CN111702754A (en) * | 2020-05-14 | 2020-09-25 | 国网安徽省电力有限公司检修分公司 | Robot obstacle avoidance trajectory planning method based on simulation learning and robot |
CN114625151A (en) * | 2022-03-10 | 2022-06-14 | 大连理工大学 | Underwater robot obstacle avoidance path planning method based on reinforcement learning |
CN114637220A (en) * | 2022-01-28 | 2022-06-17 | 中国人民解放军军事科学院国防科技创新研究院 | Semi-physical test system and method for application verification of robot operating system |
-
2023
- 2023-11-30 CN CN202311620923.2A patent/CN117325186A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2634626A1 (en) * | 2008-07-09 | 2010-01-09 | Gerard Voon | Neural links and artificial intelligence |
CN106780639A (en) * | 2017-01-20 | 2017-05-31 | 中国海洋大学 | Hash coding method based on the sparse insertion of significant characteristics and extreme learning machine |
CN111702754A (en) * | 2020-05-14 | 2020-09-25 | 国网安徽省电力有限公司检修分公司 | Robot obstacle avoidance trajectory planning method based on simulation learning and robot |
CN114637220A (en) * | 2022-01-28 | 2022-06-17 | 中国人民解放军军事科学院国防科技创新研究院 | Semi-physical test system and method for application verification of robot operating system |
CN114625151A (en) * | 2022-03-10 | 2022-06-14 | 大连理工大学 | Underwater robot obstacle avoidance path planning method based on reinforcement learning |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118143950A (en) * | 2024-04-16 | 2024-06-07 | 江苏宁昆机器人智能科技有限公司 | Industrial robot accurate operation method and system based on advanced perception |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zou et al. | A modified differential evolution algorithm for unconstrained optimization problems | |
Chen et al. | Dynamic movement primitives in latent space of time-dependent variational autoencoders | |
Wen et al. | Learning ensemble of decision trees through multifactorial genetic programming | |
Wu et al. | From tensor-network quantum states to tensorial recurrent neural networks | |
CN115862751B (en) | Quantum chemistry calculation method for updating aggregation attention mechanism based on edge features | |
Liu et al. | Telecommunication traffic prediction based on improved LSSVM | |
Zhang et al. | Robot grasping method optimization using improved deep deterministic policy gradient algorithm of deep reinforcement learning | |
CN102915407A (en) | Prediction method for three-dimensional structure of protein based on chaos bee colony algorithm | |
Broni-Bediako et al. | Evolutionary NAS with gene expression programming of cellular encoding | |
Zeng et al. | A hyperparameter adaptive genetic algorithm based on DQN | |
Johnson et al. | Learning sampling dictionaries for efficient and generalizable robot motion planning with transformers | |
CN117325186A (en) | Sampling mechanical arm path planning-based method and system | |
Yan et al. | Exkgr: Explainable multi-hop reasoning for evolving knowledge graph | |
Benitez et al. | Hierarchical parallel genetic algorithm applied to the three-dimensional HP side-chain protein folding problem | |
Yang et al. | Optimization of classification algorithm based on gene expression programming | |
Choi et al. | Recurrent DETR: Transformer-Based Object Detection for Crowded Scenes | |
Thais et al. | Equivariance Is Not All You Need: Characterizing the Utility of Equivariant Graph Neural Networks for Particle Physics Tasks | |
Garov et al. | Application and some fundamental study of GNN in forecasting | |
Lin et al. | Protein folding structure optimization based on GAPSO algorithm in the off-lattice model | |
Dianting et al. | Trajectory Optimization of Two-joint Manipulator Based on Differential Evolution Algorithm | |
Rajasekhar et al. | Cooperative micro artificial bee colony algorithm for large scale global optimization problems | |
Zhang | Application swarming intelligence in Bioinformatics: survey | |
Cofala et al. | Spatial Generation of Molecules with Transformers | |
Moskalev et al. | SE (3)-Hyena Operator for Scalable Equivariant Learning | |
Jiang et al. | A Novel Genetic Algorithm for Parameter Estimation of Sinusoidal Signals |
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 |