CN110653824A - Method for characterizing and generalizing discrete trajectory of robot based on probability model - Google Patents
Method for characterizing and generalizing discrete trajectory of robot based on probability model Download PDFInfo
- Publication number
- CN110653824A CN110653824A CN201910973457.3A CN201910973457A CN110653824A CN 110653824 A CN110653824 A CN 110653824A CN 201910973457 A CN201910973457 A CN 201910973457A CN 110653824 A CN110653824 A CN 110653824A
- Authority
- CN
- China
- Prior art keywords
- track
- discrete
- tracks
- teaching
- model
- 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
Images
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
Abstract
The invention relates to a characterization and generalization method of discrete type track of robot based on probability model, which comprises the following steps: splitting the track into multiple sections, respectively teaching each section of track, acquiring a data source of discrete track characterization, and characterizing the discrete track: modeling the robot track based on a plurality of GMMs, extracting the correlation among a plurality of sections of tracks, representing the teaching track, and generalizing the track to output: and splicing the multiple sections of tracks through GMR to realize generalized output of the tracks, so that the output tracks have smoothness. The teaching process is simplified and the operability is strong; smooth splicing can be carried out on each track based on the time information; by learning the multitask constraint relation of the multiple mechanical arms, the multiple mechanical arms of the robot can cooperatively complete multitask.
Description
Technical Field
The invention relates to a method for characterizing and generalizing discrete type tracks of a robot based on a probability model.
Background
Teaching learning allows a robot to learn how a human performs a smart operation task in an unknown environment and to generate a robot trajectory meeting the requirements in a new environment and task goals. The trajectory generation strategy based on teaching learning can fully extract the characteristics of the teaching trajectory and generate the robot trajectory with certain generalization.
Discrete tracks are common in human life, such as Chinese character writing. But the teaching learning field has less research on feature extraction of discrete trajectories. The current research on the characterization and generalization of the discrete trajectory of the robot mainly has the following defects:
(1) and the teaching track is complicated and tedious. The acquisition of the discrete teaching data is basically obtained by continuous operation at present. When the learned track is complex, the teaching strategy is obviously complicated, a great amount of time and cost are needed in the teaching process, and a certain skill is also needed to continuously design the track.
(2) And lack of corresponding trajectory stitching strategies. Different from a continuous track, a discrete track is composed of multiple sections of tracks, certain space constraint and time constraint exist among the sections of tracks, corresponding strategies are lacked in the current researches on the characterization and generalization of the discrete track to splice the multiple discrete tracks, and the final output track can not have good smoothness while the related constraint of the original track is met.
Disclosure of Invention
The invention aims to provide a method for characterizing and generalizing discrete type tracks of a robot based on a probability model.
In order to achieve the purpose, the invention adopts the technical scheme that:
a method for characterizing and generalizing discrete trajectory of robot based on probability model includes:
(1) teaching of discrete trajectory: splitting the track into a plurality of sections, respectively teaching each section of track to obtain a data source of discrete track representation,
(2) and characterizing the discrete type track: modeling the robot track based on a plurality of Gaussian Mixture Models (GMMs), extracting the correlation among a plurality of sections of tracks, representing the teaching track,
(3) and outputting track generalization: and splicing the multiple tracks through Gaussian Mixture Regression (GMR) to realize generalized output of the tracks, so that the output tracks have smoothness.
Preferably, the data are clustered by using a k-means clustering algorithm (k-means), the class of the teaching data is divided, and the data of the Gaussian Mixture Model (GMM) are learned by using a maximum expectation algorithm (EM algorithm).
Further preferably, the learning of the data of the Gaussian Mixture Model (GMM) using the maximum expectation algorithm (EM algorithm) is performed cyclically using the estimation step (E-step) and the maximization step (M-step) until the parameters converge.
Further preferably, in the estimating step (E-step), it includes:
(1) classifying and dividing the sampling data,
(2) and the probability P (yt, gamma t | mu, Σ, pi) of generating a sample is determined for each class (K1, 2 … K),
(3) and solving a probability Q function generated by the sampling data.
Further preferably, the Q-function is maximized in a maximization step (M-step) to optimize parameters of the Gaussian Mixture Model (GMM).
Preferably, in (1): and acquiring a data source of discrete track representation by using a dragging teaching strategy.
Preferably, in (2): and (3) utilizing Matlab programming to realize a Gaussian Mixture Model (GMM) to characterize the teaching track.
Preferably, when modeling the robot trajectory by a Gaussian Mixture Model (GMM), a probability model is used to extract the correlation between the multiple segments of the trajectory.
Preferably, in (3): when the track is generalized and outputted by the Gaussian Mixture Regression (GMR), the generalized output is saved as a mat file as a desired track of the control system.
Preferably, the method further comprises (4) utilizing the generalized output trajectory as a desired trajectory for the control system, such that the robot trajectory tracks the desired trajectory
Further preferably, the robot performs the trajectory learning of the control system by multi-robot multi-task coordination.
Due to the application of the technical scheme, compared with the prior art, the invention has the following advantages and effects:
1. discretization and subsection teaching are carried out on the discrete track of the complex robot, so that the teaching process is simplified and the operability is strong;
2. representing multiple discrete tracks by adopting multiple GMMs, introducing time dimension information among the tracks, and performing smooth splicing on the tracks based on the time information when the GMRs can be finally used for track generalization output;
3. by learning the multitask constraint relation of the multiple mechanical arms, the multiple mechanical arms of the robot can cooperatively complete multitask.
Drawings
FIG. 1 is a schematic flow chart of the method of this embodiment;
FIG. 2 is a diagram illustrating GMM model parameter learning in this embodiment;
FIG. 3 is a graph of the trajectory teaching, GMM characterization, and GMR generalized output trajectory for the present embodiment;
FIG. 4 is a representation and generalized trajectory diagram of a trajectory "typing" continuous teaching strategy;
FIG. 5 is a representation and generalized trajectory diagram of a discrete teaching strategy for trajectory typing;
FIG. 6 is a diagram of the trajectory output of multi-robot multi-task coordination.
Detailed Description
The invention is further described below with reference to the accompanying drawings and embodiments:
as shown in the figure: a method for characterizing and generalizing discrete trajectory of robot based on probability model includes:
(1) teaching of discrete trajectory:
the source of the teaching data is obtained by using a strategy of dragging teaching, and the teaching data is firstly expressed: for two-dimensional teaching data, this is expressed herein as:
wherein, yi,s,yi,tThe spatial information and the time information of the teaching trajectory are respectively shown, and T shows the number of teaching points in the teaching trajectory.
(2) And characterizing discrete tracks and learning parameters:
for a multidimensional teaching variable y, the modeled GMM is:
wherein p (y) represents a probability density function, N (y, μ)k,∑k) Is expressed in μkIs mean value, ΣkIs a gaussian probability density function of the covariance matrix.
Compared with the parameter estimation of a Gaussian model, the parameter estimation of the Gaussian mixture model is more complicated, and the main reason is that the existence of hidden variables cannot utilize a maximum likelihood estimation method to obtain the parameters of the model. For the teaching sample set Y ═ (Y)1,y2…yT) By an implicit variable gammat,KCan be expanded into full data:
(yt,γt,1,γt,2…γt,K),t=1,2...T (3)
if ytFrom class 1 sampling, then there is γt,1=1,γt,2=0…γt,KIs represented by (y) 0t,1,0,…0)。
The likelihood function for the complete data is:
the log-likelihood function for the full data is:
the Q function is defined as follows:
wherein, E (γ)t,K|yt,μi,∑i,πi) Is an estimate of γ:
the Q function is derived and its derivative is 0, which can be:
whereinRespectively represent the (i +1) th iteration, the mean of the kth class, the covariance matrix, and the occupied weight.
(3) And outputting track generalization:
and after the teaching track is subjected to GMM representation coding, utilizing GMR to output the track.
Data point y for the teach path [ y ═ yI,yo]First, the distribution P (y) of the teaching data points is calculated using a probability modelI,yo) Modeled as GMM, followed by computation of the condition variable (y) by GMRo|yI) Is desired E (y)o|yI) And covariance Cov (y)o|yI) Mixing E (y)o|yI) As a generalized output data point, in Cov (y)o|yI) And generating a motion track with smoothness under the constraint.
For a dataset of T D-dimensional teach data points, the GMM is modeled as follows:
wherein, pikIs the prior probability of the model, N (y, μ)k,∑k) Is measured in mukAs a mean value, by ∑kIs a gaussian distribution of variance and has:
at a given yIAnd the k-th Gaussian distribution, the condition variable (y)o|yIK) also follows a gaussian distribution, i.e.:
(yo|yI,k)~N(μ′k,∑′k) (13)
wherein, muk,∑kRespectively as follows:
for the entire GMM, then there is (y)o|yI) Satisfies the following conditions:
wherein h iskSatisfies the following conditions:
from this, the condition variable (y)o|yI) The mean μ and covariance ∑ of are:
(4) and designing a control system:
and designing a control system on a working space, and carrying out a multi-task collaborative track learning strategy by a plurality of mechanical arms of the robot. The teaching data set for GMM may be represented as y ═ yI,yo]Wherein y isIAnd yoRespectively a query vector and a vector to be encoded. In GMM-based robot trajectory characterization learning, the teaching data set is y ═ yI t,yo s]That is, the query vector is time information, and the vector to be encoded is spatial information of the track. For multi-robot multi-task collaborative track learning, a two-dimensional space vector (y) of a certain robot is usedI s1,yI s2) As a query vector, the two-dimensional space vector (y) of the remaining mechanical armsO s1,yO s2,yO s3,yO s4…yO s2n) And as a vector to be coded, performing the characterization and learning of the track. For example, for the two-robot-arm two-task learning, it is necessary to construct a 4-dimensional (2 × 2) robot teaching dataset y ═ yI s1,yI s2,yO s1,yO s2) I.e. yI=(yI s1,yI s2),y=(yO s1,yO s2). During the track generalization output process, yIFor query points, P (y) is estimated using GMRo|yI) For the rest two-dimensional space information yOAnd outputting to realize the double-task cooperation of the double mechanical arms.
The above embodiments are merely illustrative of the technical ideas and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the contents of the present invention and implement the present invention, and not to limit the protection scope of the present invention. All equivalent changes and modifications made according to the spirit of the present invention should be covered within the protection scope of the present invention.
Claims (10)
1. A characterization and generalization method of discrete trajectory of robot based on probability model is characterized in that: the method comprises the following steps:
(1) teaching of discrete trajectory: splitting the track into a plurality of sections, respectively teaching each section of track to obtain a data source of discrete track representation,
(2) and characterizing the discrete type track: modeling robot tracks through a plurality of Gaussian Mixture Models (GMMs), extracting the correlation among a plurality of sections of tracks, representing teaching tracks,
(3) and outputting track generalization: and splicing the multiple tracks through Gaussian Mixture Regression (GMR) to realize generalized output of the tracks.
2. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 1, wherein: and clustering the data by using a k-means clustering algorithm (k-means), dividing the class of the teaching data, and learning the data of the Gaussian Mixture Model (GMM) by using a maximum expectation algorithm (EM algorithm).
3. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 2, wherein: when learning data of a Gaussian Mixture Model (GMM) by using a maximum expectation algorithm (EM algorithm), the estimation step (E-step) and the maximization step (M-step) are used for circulating until parameters are converged.
4. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 3, wherein: in the estimating step (E-step), it comprises:
(1) classifying and dividing the sampling data,
(2) and the probability P (yt, gamma t | mu, Σ, pi) of generating a sample is determined for each class (K1, 2 … K),
(3) and solving a probability Q function generated by the sampling data.
5. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 4, wherein: the Q function is maximized in a maximization step (M-step) to optimize the parameters of the Gaussian Mixture Model (GMM).
6. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 1, wherein: in (1): and acquiring a data source of discrete track representation by using a dragging teaching strategy.
7. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 1, wherein: in (2): and (3) utilizing Matlab programming to realize a Gaussian Mixture Model (GMM) to characterize the teaching track.
8. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 1, wherein: when the robot track is modeled by a Gaussian Mixture Model (GMM), a probability model is used for extracting the correlation among multiple sections of tracks.
9. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 1, wherein: in (3): when the track is generalized and outputted by the Gaussian Mixture Regression (GMR), the generalized output is saved as a mat file as a desired track of the control system.
10. The probabilistic model-based characterization and generalization method for discrete trajectories of robots according to claim 1, wherein: it also includes (4), control system design: and utilizing the track of the generalized output as a desired track of the control system, so that the robot track tracks the desired track.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910682654X | 2019-07-26 | ||
CN201910682654 | 2019-07-26 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110653824A true CN110653824A (en) | 2020-01-07 |
CN110653824B CN110653824B (en) | 2021-06-04 |
Family
ID=69040837
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910973457.3A Active CN110653824B (en) | 2019-07-26 | 2019-10-14 | Method for characterizing and generalizing discrete trajectory of robot based on probability model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110653824B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111251277A (en) * | 2020-01-31 | 2020-06-09 | 武汉大学 | Human-computer collaboration tool submission system and method based on teaching learning |
CN111424380A (en) * | 2020-03-31 | 2020-07-17 | 山东大学 | Robot sewing system and method based on skill learning and generalization |
CN111859297A (en) * | 2020-07-20 | 2020-10-30 | 上海交通大学 | Retort loading track extraction method based on Gaussian mixture model |
CN114227688A (en) * | 2021-12-29 | 2022-03-25 | 同济大学 | Teaching trajectory learning method based on curve registration |
CN115026842A (en) * | 2022-08-11 | 2022-09-09 | 深圳市创智机器人有限公司 | Teaching track processing method and device, terminal device and storage medium |
WO2023142215A1 (en) * | 2022-01-27 | 2023-08-03 | 苏州大学 | Method for automatically picking up nanowires by micro-nano operation robot on basis of dynamic motion primitives |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2077970A1 (en) * | 1991-11-19 | 1993-05-20 | Daniel P. Huttenlocher | Optical word recognition by examination of word shape |
CN104392241A (en) * | 2014-11-05 | 2015-03-04 | 电子科技大学 | Mixed regression-based head pose estimation method |
CN105760891A (en) * | 2016-03-02 | 2016-07-13 | 上海源庐加佳信息科技有限公司 | Chinese character verification code recognition method |
CN105956601A (en) * | 2016-04-15 | 2016-09-21 | 北京工业大学 | Robot Chinese character writing learning method based on track imitation |
CN108921935A (en) * | 2018-06-05 | 2018-11-30 | 中国人民解放军军事科学院国防科技创新研究院 | A kind of extraterrestrial target method for reconstructing based on acceleration gauss hybrid models |
-
2019
- 2019-10-14 CN CN201910973457.3A patent/CN110653824B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2077970A1 (en) * | 1991-11-19 | 1993-05-20 | Daniel P. Huttenlocher | Optical word recognition by examination of word shape |
CN104392241A (en) * | 2014-11-05 | 2015-03-04 | 电子科技大学 | Mixed regression-based head pose estimation method |
CN105760891A (en) * | 2016-03-02 | 2016-07-13 | 上海源庐加佳信息科技有限公司 | Chinese character verification code recognition method |
CN105956601A (en) * | 2016-04-15 | 2016-09-21 | 北京工业大学 | Robot Chinese character writing learning method based on track imitation |
CN108921935A (en) * | 2018-06-05 | 2018-11-30 | 中国人民解放军军事科学院国防科技创新研究院 | A kind of extraterrestrial target method for reconstructing based on acceleration gauss hybrid models |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111251277A (en) * | 2020-01-31 | 2020-06-09 | 武汉大学 | Human-computer collaboration tool submission system and method based on teaching learning |
CN111251277B (en) * | 2020-01-31 | 2021-09-03 | 武汉大学 | Human-computer collaboration tool submission system and method based on teaching learning |
CN111424380A (en) * | 2020-03-31 | 2020-07-17 | 山东大学 | Robot sewing system and method based on skill learning and generalization |
CN111859297A (en) * | 2020-07-20 | 2020-10-30 | 上海交通大学 | Retort loading track extraction method based on Gaussian mixture model |
CN114227688A (en) * | 2021-12-29 | 2022-03-25 | 同济大学 | Teaching trajectory learning method based on curve registration |
CN114227688B (en) * | 2021-12-29 | 2023-08-04 | 同济大学 | Teaching track learning method based on curve registration |
WO2023142215A1 (en) * | 2022-01-27 | 2023-08-03 | 苏州大学 | Method for automatically picking up nanowires by micro-nano operation robot on basis of dynamic motion primitives |
CN115026842A (en) * | 2022-08-11 | 2022-09-09 | 深圳市创智机器人有限公司 | Teaching track processing method and device, terminal device and storage medium |
CN115026842B (en) * | 2022-08-11 | 2022-12-02 | 深圳市创智机器人有限公司 | Teaching track processing method and device, terminal device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110653824B (en) | 2021-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110653824B (en) | Method for characterizing and generalizing discrete trajectory of robot based on probability model | |
Sener et al. | Unsupervised learning and segmentation of complex activities from video | |
CN105956601B (en) | A kind of robot Chinese writing and learning method based on Track Imitation | |
Constantinopoulos et al. | Bayesian feature and model selection for Gaussian mixture models | |
CN112605973B (en) | Robot motor skill learning method and system | |
Ontanón et al. | A dynamic-bayesian network framework for modeling and evaluating learning from observation | |
Lin et al. | A deep clustering algorithm based on gaussian mixture model | |
Lippi et al. | Enabling visual action planning for object manipulation through latent space roadmap | |
Chen et al. | PaDGAN: A generative adversarial network for performance augmented diverse designs | |
García-Cuesta et al. | User modeling: Through statistical analysis and subspace learning | |
Pal | Generative adversarial network-based cross-project fault prediction | |
KR101329642B1 (en) | Modeling method for learning task skill and robot using thereof | |
Okada et al. | Incremental learning of gestures for human–robot interaction | |
Sun et al. | Expressing diverse human driving behavior with probabilistic rewards and online inference | |
Vinokurov et al. | Unsurpervised learning in hybrid cognitive architectures | |
Yin et al. | Learning cost function and trajectory for robotic writing motion | |
Zhu et al. | Fast Adaptive Character Animation Synthesis Based on Greedy Algorithm | |
Pan et al. | Unsupervised hierarchical modeling of locomotion styles | |
Kulić et al. | Incremental learning of full body motion primitives | |
Fan et al. | Unsupervised modeling and feature selection of sequential spherical data through nonparametric hidden Markov models | |
Hajimirsadeghi et al. | Conceptual imitation learning: An application to human-robot interaction | |
Marttinen et al. | Bayesian clustering of fuzzy feature vectors using a quasi-likelihood approach | |
CN112199287A (en) | Cross-project software defect prediction method based on enhanced hybrid expert model | |
Hajimirsadeghi et al. | Conceptual imitation learning in a human-robot interaction paradigm | |
Suder et al. | Bayesian Transfer Learning |
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 |