CN115700414A - Robot motion error compensation method - Google Patents
Robot motion error compensation method Download PDFInfo
- Publication number
- CN115700414A CN115700414A CN202211382761.9A CN202211382761A CN115700414A CN 115700414 A CN115700414 A CN 115700414A CN 202211382761 A CN202211382761 A CN 202211382761A CN 115700414 A CN115700414 A CN 115700414A
- Authority
- CN
- China
- Prior art keywords
- robot
- algorithm
- pose
- error
- motion error
- 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
Images
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses a robot motion error compensation method which is characterized by comprising the following steps: step one, training a robot motion error prediction model by using an acquired error data set; correcting the robot motion planning output by using the trained motion error prediction model; the motion error prediction model is as follows: the ELM network model, the training method of the ELM network model adopts a novel metaheuristic optimization algorithm, namely the improved mayday algorithm MMA; the MMA-trained ELM network model is abbreviated as MMA-ELM model. The robot motion error compensation method provided by the invention has the advantages of simple structure, high prediction precision, stable performance, high training speed and the like during robot motion error compensation.
Description
Technical Field
The invention relates to the technical field of robot motion error compensation, in particular to a robot motion error compensation method.
Background
For complex robot systems with multiple connecting rods and multiple joints, such as serial robots and parallel robots, unavoidable motion errors exist between the actual motion trajectory of the tail end of the robot execution and the theoretically planned motion trajectory due to factors such as connecting rod deformation, joint gaps, machining errors and assembly errors, the motion errors can affect the accuracy of the motion of the robot, and further the robot is difficult to accurately complete set tasks, and even serious consequences are caused in actual operation scenes.
The fundamental reason for causing the robot motion error is that the established theoretical kinematics model and the actual kinematics model of the robot have differences, and the differences are caused by error sources such as the connecting rod deformation, the joint clearance, the processing error and the assembly error; among many error sources, some factors can be represented by constant parameterization and identified and compensated by a modeling mode, which is called as geometric errors; some factors cannot be compensated by a traditional modeling method, and even cannot be accurately represented by using parameters, which are collectively referred to as non-geometric errors.
The relationship between the actual motion pose (position and attitude) of the robot and the theoretical motion pose can be expressed by the following equation:
T a ＝T t +ΔT＝T t +ΔT g +ΔT ng (1)
t in the above formula (1) a ＝[P xa ,P ya ,P za ,α a ,β a ,γ a ] T Representing the robot to execute the terminal actual motion pose matrix, where [ P xa ,P ya ,P za ]Is the robot execution end coordinate system origin O f In a reference coordinate system sigma O 0 Position coordinates of (1) [ alpha ] a ,β a ,γ a ]Is a robot executing end coordinate system sigmaO f Relative to a reference coordinate system Σ O 0 The attitude euler angle of (a); t is a unit of t The matrix is a theoretical motion pose matrix of the robot executing the tail end, and the delta T is a total error matrix of the theoretical motion and the actual motion of the robot, and comprises a geometric error matrix delta T g And a non-geometric error matrix Δ T ng (ii) a In order to realize compensation of motion errors, the core lies in establishing a prediction model of errors, namely establishing a mapping relation shown as a formula (2), and predicting the pose error delta T of the robot execution end through a given motion input joint angle position theta. After the error prediction model is obtained, the angle position theta of the motion input joint of the robot can be corrected, so that the pose output of the execution tail end of the robot is corrected, and the finally measured actual pose error is reduced.
ΔT＝M(θ) (2)
In the prior art, error compensation algorithms for robots are generally classified into model-based algorithms and model-free algorithms; the traditional model-based compensation algorithm generally utilizes a structural parameter and a kinematic model to establish a geometric error model, and then can identify the geometric error parameter through various identification methods, such as a least square method, a maximum likelihood estimation method, a meta-heuristic algorithm, an extended Kalman filtering method and the like; for non-geometric type error compensation, these errors cannot be identified as constants using the model-based method described above because their causes are complex and time-varying. Therefore, model-free algorithms can be used to compensate for such errors, i.e. using the data set to directly map the motion error to the robot input. The artificial neural network has the advantages of high adaptability, strong learning ability, easy use and the like, and is a very popular model-free error compensation method at present.
In the prior art, in patent CN115179289A, a robot rigid-flexible coupling error model is established on a robot calibration model by a linear superposition principle, so that the accuracy of the model is improved; the patent CN115026819A provides a robot calibration method based on the FIS theory, and researches on error modeling and identification are carried out based on a finite instantaneous momentum theory system, so that a robot error model meeting completeness, continuity and minimum conditions is established; patent CN114918920A proposes an industrial robot calibration method based on a neural network and a distance error model, which avoids a transformation error of a coordinate system in a calibration process by establishing a model relationship between a robot terminal distance error and a robot kinematic parameter error, and considers a connecting rod error and a joint flexibility error, and fits a residual error through the neural network after identifying an error parameter, so as to improve the absolute positioning accuracy of the industrial robot; the patent CN114820813A proposes a hand-eye calibration method based on a BP neural network with SVD supervision, and the algorithm firstly pre-trains the BP neural network, and then optimizes the neural network by using the SVD algorithm to improve the accuracy and robustness of the algorithm. The patent CN202111608435.0 proposes a robot calibration method combining geometric errors and non-geometric errors, in which the algorithm firstly uses a robot kinematic model and a singular value decomposition method to identify geometric error parameters, and then uses a particle swarm algorithm to compensate the non-geometric errors; patent CN113043271B proposes an industrial robot calibration compensation method based on a longicorn whisker algorithm, which solves the accurate parameters of the kinematics model of the industrial robot by establishing a robot kinematics model and then utilizing the longicorn whisker algorithm to iterate.
In the prior art, a common error compensation algorithm based on a model can only compensate geometric errors of robot motion, but cannot compensate complex non-geometric errors with time-varying property. In the model-free error compensation algorithm, the traditional artificial neural network is usually trained by using a back propagation algorithm, so that the application of the artificial neural network is limited by the defects of long training time, low convergence rate, poor generalization and the like. In order to solve the problem, people use meta-heuristic algorithm to train the neural network, for example, a novel calibration method based on artificial neural network combined with butterfly and flower pollination algorithm is proposed, and people combine the neural network with particle swarm optimization algorithm to calibrate the industrial robot. A position error prediction model has also been established using genetic swarm optimization and deep neural networks. However, these methods still have problems of not fast convergence speed, requiring long training time, and being prone to trap in locally optimal traps.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a robot motion error compensation method, and aims to solve the problems.
In order to achieve the purpose, the invention provides the following technical scheme:
a robot motion error compensation method comprises two steps: firstly, training a robot motion error prediction model by using an acquired error data set, and secondly, correcting robot motion planning output by using the trained motion error prediction model;
preferably, the error data set in the first step includes a pose error data set
And driving the joint target angular position data set
(N is the number of samples contained in the data set);
preferably, the pose error data set
The obtaining method is as follows: firstly, in a reachable working space at the execution end of the robot, randomly screening N target point positions to form a target pose data set
Then controlling the execution tail end of the robot to reach the point positions, actually measuring the position and pose of each point position by using external measuring equipment, and forming an actually measured pose data set by the actually measured values
The difference between the actual measurement positions and the target positions of the robot execution tail end at the point positions is the motion error of the robot, and a position error data set is formed by the actual measurement positions and the target positions
The label set is used as a label set for training a motion error prediction model;
preferably, the data set of angular positions of the target of the driving joint
The obtaining method is as follows: object pose data set
Input to the ideal inverse kinematics model f of the robot - (T ft ) In order to obtain
Preferably, the motion error prediction model in the step one refers to an Extreme Learning Machine (ELM) network model, and the training method of the model adopts a novel metaheuristic optimization Algorithm, namely a Modified Mayfly Algorithm (MMA), where the ELM network model trained by MMA is referred to as MMA-ELM model for short;
preferably, the improved mayflies Algorithm is an improvement of the initialization strategy of the original Mayflies Algorithm (MA) by using the improved Tent chaotic map and the reverse learning theory to improve the initial conditions and convergence rate of the mayflies Algorithm; in addition, the Cauchy variation theory is adopted to improve the position updating strategy of the mayfly algorithm so as to reduce the probability of the algorithm falling into the locally optimal traps;
preferably, the specific process of correcting the robot motion planning output by using the trained motion error prediction model in the step two is as follows: a new robot execution end target point position is planned in the working space, and the position and posture matrix of the point position is T ft It is input into an inverse kinematics model f - (T ft ) Then the target angle position theta of the robot driving joint can be obtained t (ii) a The theta being t The method is not directly used for driving the robot, but the robot execution end pose error delta T under the instruction is estimated through a trained ELM network f (ii) a Then, the original target pose T is determined ft And estimate the pose error delta T f Adding to obtain a new target pose T fc ＝T ft +ΔT f (ii) a Finally, T is added fc Input into an inverse kinematics model f - (T fc ) The compensated angle position theta of the driving joint can be obtained c ，θ c It is the final angle command value for driving the robot.
Advantageous effects
Compared with the prior art, the invention has the following beneficial effects:
the robot motion error compensation method provided by the invention predicts the motion error of the execution tail end of the robot based on the MMA-ELM network model optimized by the improved mayfly algorithm, retains the advantages of simple structure, high training speed and the like of the ELM network, and overcomes the defects of unstable performance, difficulty in finding an optimal solution and the like of the ELM network; therefore, the method provided by the invention has the advantages of simple structure, high prediction precision, stable performance, high training speed and the like when the robot motion error is compensated.
Drawings
Fig. 1 is a flow chart of a robot motion error compensation method according to the present invention;
fig. 2 is a diagram of an ELM network model architecture.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, a method for compensating a motion error of a robot includes the following steps:
firstly, training a robot motion error prediction model by using an acquired error data set;
and step two, correcting the robot motion planning output by using the trained motion error prediction model.
The error data set in the step one comprises a pose error data set
And driving the joint target angular position data set
Where N is the number of samples contained in the dataset.
Pose error dataset of
The obtaining method is as follows:
1. in the reachable working space of the execution tail end of the robot, N target point positions are randomly screened out to form a target pose data set
2. Controlling the execution tail end of the robot to reach the point positions, actually measuring the position and pose of each point position by using external measuring equipment, and forming an actually measured position and pose data set by the actually measured values
3. The difference between the actual pose and the target pose of the robot's execution end at these points is the robot's motion error, which constitutes a pose error dataset
As a set of labels for training a motion error prediction model.
Driving joint target angular position data set of
The obtaining method is as follows:
object pose data set
Ideal inverse kinematics model f input to a robot - (T ft ) In order to obtain
The motion error prediction model in the first step is as follows:
the ELM network model, the training method of the ELM network model adopts a novel metaheuristic optimization algorithm, namely the improved mayday algorithm MMA;
the MMA-ELM model is abbreviated as the MMA-ELM model.
The improved mayflies algorithm MMA improves the initialization strategy of the original Mayflies Algorithm (MA) by using an improved Tent chaotic map and the inverse learning theory to improve the initial conditions and convergence speed of the mayflies algorithm;
simultaneously, the Cauchy variation theory is employed to improve the position updating strategy of the mayflies so as to reduce the probability of the algorithm falling into locally optimal traps.
The specific process of the second step is as follows:
1. planning a new robot execution end target point position in the working space, wherein the position and posture matrix of the point position is T ft Inputting it into an inverse kinematics model f - (T ft ) Then the target angle position theta of the robot driving joint can be obtained t ；
The theta being t The method is not directly used for driving the robot, but predicts the robot execution end pose error delta T under the instruction through a trained ELM network f ；
2. The original target pose T ft And the estimated pose error delta T f After addition, a new target pose T is obtained fc ＝T ft +ΔT f ；
3. Will T fc Input into an inverse kinematics model f - (T fc ) Can obtain the compensated angle position theta of the driving joint c ，θ c Is the most used for driving the robotThe final angle command value.
The MMA-ELM model is specified below:
the neural network structure design based on the ELM comprises the following steps:
the ELM is similar to the traditional neural network in network structure, is a single hidden layer feedforward neural network, and has the characteristics of simple network structure, good model generalization capability, high training speed and the like.
As shown in fig. 2, the ELM network designed by the present invention includes 3 network layers: an input layer, a hidden layer, and an output layer; taking a 6-degree-of-freedom tandem articulated robot as an example (the method of other robots is similar), the input layer of the ELM network model is set with 6 nodes, and the nodes correspond to 6 driving joint angle values θ = [ theta ], (the method is similar for other robots) 1 ,θ 2 ,θ 3 ,θ 4 ,θ 5 ,θ 6 ] T (ii) a The output layer of the ELM network also has 6 nodes which respectively correspond to the estimated motion pose error delta T = [ delta P ] of the execution end of the robot x ,ΔP y ,ΔP z ,Δα,Δβ,Δγ] T (ii) a And the number of hidden layer nodes of the ELM network is n B The specific values are adjusted according to actual conditions. Based on fig. 2, the Δ T and θ mapping L () can be expressed by the following expression (3):
ΔT＝L(θ)＝V·g(W·θ+B) (3)
in the formula (I), the compound is shown in the specification,
represents a matrix of input-layer weight coefficients,
a matrix of hidden layer bias coefficients is represented,
represents the output layer weight coefficient matrix, and g () represents the activation function, where the usual Sigmoid activation function is selected, whose expression is shown in the following formula (4):
the ELM network is trained by utilizing the data set, so that the optimal W, B and V coefficient matrixes are obtained, and the estimation error of the network on the delta T is minimized; for a traditional BP (Back Propagation) neural network, it is necessary to adjust the weighting coefficients of the input layer, the hidden layer and the output layer by collecting the output error of the network to obtain the optimal parameter combination, and this process consumes a lot of time and computational power; the innovation of ELM is that: the input layer W and the hidden layer B are generated in a random mode, and adjustment is not needed after generation; the output layer V also does not need to be adjusted in an iterative manner, but is obtained once by solving a system of equations. Therefore, the training speed of ELM is much faster than that of BP neural network with the same structure. Training set given a set of numbers
And
rear (N) tr As the data amount of the training set), V can be obtained by the following equation (5):
in the formula (I), the compound is shown in the specification,
and
is a randomly generated coefficient matrix in which
The value range of the elements in the formula is [ -1, +1 [)]，
The value range of the middle element is [0, +1 ]]To do so
Is N tr The columns are the same
The constituent matrices, the symbol + represents the Moore-Penrose generalized inverse of the matrix.
Since the ELM network parameters are obtained in an iterative optimization mode, the method has the advantage of high training speed; however, since the input layer W and the hidden layer B are generated randomly, the ELM has certain limitations: the random initialization condition is seriously relied on, and the performance is unstable; it is difficult to find a global optimal solution; in order to overcome the defects and keep the performance advantage of the ELM part, a novel MMA optimization algorithm is adopted to obtain W and B coefficient matrixes, and the V coefficient matrix is directly obtained through a Moore-Penrose generalized inverse shown in a formula (5).
Optimization algorithm design based on original MA
The MA algorithm is a novel metaheuristic optimization algorithm developed by Konstantinos Zervoudakis et al that is inspirational from the flight and mating behavior of both females and males. In a collection of mayflies, the position of each mayflies in a defined search space represents a solution, the purpose of the algorithm is to find the globally optimal position which maximizes the fitness value by simulating the social behavior of the mayflies. For optimization of ELM network structure parameters, the position of each mayfly can be used as one multidimensional vector x = (x) 1 ,…,x i ,…,x D ) To indicate that x consists of all elements in the W and B coefficient matrices, the dimension of the position vector D = n I ×n B +n B In the formula n I Number of nodes, n, for input layer of ELM network B The number of nodes of the hidden layer. To evaluate the goodness of performance per mayfly position, an objective function was designed to calculate fitness values for the various positions, as shown in equation (6) below:
in the formula, lb kj K-th position error label in presentation verification set
Of the jth node element, n o Is the number of nodes of the output layer of the ELM network, N va The number of samples in the verification set is verified; when the input of the ELM network is verification centralized
The j-th node element of the network output value is in ot kj Represents; the purpose of the MA algorithm is to find a set of global optimal positions
Making the above-mentioned target function have minimum fitness value
The MA algorithm mainly includes the following operations: position initialization, mayflies, male mayflies, and female and male mayflies. To describe these courses of operation, the positions of the male mayfly population are shown as
The position of the female mayfly population is shown as
In the formula N ma And N fm Denotes the number of male and female mayfly colonies, respectively, usually N ma ＝N fm (ii) a The positions of male and female mayflies can be initialized by equations (7) and (8):
in the above two equations, rand () is a function that generates uniform random numbers and limits the numbers generated by rand () to [ x [ ] min ,x max ]In the range of [ x ] herein min ,x max ]＝[-1,+1]。
Each mayflies continuously adjust their position during the iteration, with the change in position being influenced by the individual's historical optimum position
And the optimal position of the whole population
The double effect of (c). The location updating strategy differs between male and female dayflies: the male mayflies generally gather together and dance in arms through their own unique patterns of motion to attract female mayflies to mate, so the position updating of male mayflies is affected by neighboring individuals and the position updating speed is not too fast; while female dayflies do not do their myxeme, they rank according to the fitness value of their location, flying directly to the location of the male dayflies that match themselves in rank. The position updating pattern of male dayflies at each iteration cycle t is shown in the following formula (9):
in the formula (I), the compound is shown in the specification,
represents the position updating speed of the mth male mayflies in the ith dimension, and the expression thereof is shown in the following formula (10):
in which dc is a dancing coefficient with gradual attenuation, dc t+1 ＝dc t ·dc damp (0＜dc damp < 1 is a damping coefficient), a 1 And a 2 For a positive coefficient of attraction, β is the visibility coefficient of dayflies, g ∈ (0, 1)]In order to be the gravity coefficient,
the current position of mayflies and the historical optimum position of the individual
The distance between the two or more of the two or more,
the current position and the best position of the whole colony for dayflies
Of the distance of (c).
The position updating pattern of female dayflies at each iteration cycle t is as shown in the following formula (11):
in the formula (I), the compound is shown in the specification,
represents the position update speed of the mth female mayflies in the ith dimension, and the expression thereof is shown in the following formula (12):
in the formula, a 3 For positive attraction coefficients, fl is a random walk coefficient that decays gradually, fl t+1 ＝fl t ·fl damp (0＜fl damp < 1 is the attenuation coefficient), and rand is [ -1, + 1)]A random number within the range of one,
is the distance between male dayflies and female dayflies.
After the single position updates of the male and female dayflies, the fitness values for the various positions are calculated by equation (6) and sorted according to the fitness value, the mating female and male dayflies are mated and each pair of dayflies can propagate two offspring dayflies, the offspring dayflies propagated can be obtained by the following equation (13):
in the formula, N off Represents the number of offspring dayflies, L = rand (0, 1) is a random number; the initial speed of offspring dayflies was set to 0; in order to control the number of mayflies, the offspring mayflies which propagate also have their fitness values calculated and the low performance mayflies are replaced with high performance offspring mayflies; the specific working mechanism of the MA algorithm can refer to the original paper, and the basic flow thereof includes the following steps:
initializing the male and female dayfly positions and respective parameters;
the fitness values of the individual mayflies are calculated and ranked to screen out the historical best position of each mayfly individual
And the best position of the whole group
Updating the female dayfly positions and the male dayfly positions, and propagating offspring dayflies by mating;
calculating and updating the fitness value of each mayfly position
And
and the replacement of new and old mayflies is completed;
and D, judging whether the stop condition is met, if so, exiting the iteration, and otherwise, repeating the steps c) to e).
Improved strategy for MA algorithm
The MA algorithm combines the main advantages of a group intelligent algorithm and an evolutionary algorithm, has stronger optimizing capability and faster convergence rate, but has the following defects: firstly, an initial population of the MA algorithm is generated by a pure random strategy shown in formulas (7) and (8), when the random individual position is near the optimal solution, the algorithm can be converged quickly, and when all the individual initial positions are far away from the optimal position, the convergence time of the algorithm is prolonged and even falls into a local optimal trap; randomly initializing the nonuniformity problem of population individuals, so that the convergence speed of the algorithm is difficult to estimate; in addition, although the MA algorithm can enhance the search capability of the algorithm and improve the search accuracy of the algorithm by the breeding method shown in equation (13), the generation position of the offspring population is limited to the vicinity of the original population position. At the later stage of the algorithm, the MA algorithm still has a higher risk of trapping the locally optimal trap as the search range is narrowed. To improve the above problem, the MA algorithm is improved in two ways.
(1) MA population initialization strategy improvement based on improved Tent mapping and reverse learning theory
Aiming at the problem of population initialization, the invention adopts Tent chaotic mapping with high randomness and high ergodicity to replace random functions to generate the initial positions of the mayfly populations so as to improve the diversity of the initial populations and improve the global searching capability of the algorithm. On the basis, a reverse learning theory is utilized again to perform primary optimization on the initial population, so that the convergence success rate and the convergence speed of the algorithm are improved. Tent chaotic map has excellent traversal uniformity, but it has problems of small period and unstable period point, so that the present document adopts an improved Tent map whose expression is shown in the following formula (14):
in the formula, z i Representing the value of the chaos variable, a 4 E (0, 1) is a constant, usually chosen as a 4 =0.5; due to Tent mappingThe generated values are not necessarily within the range defining the mayfly population positions, and therefore, it is also necessary to carry them to a set position solution space as shown in the following formula (15):
x mi ＝x min +(x max -x min )z i (15)
an inverse learning-based learning (OBL) theory is proposed by Tizhoosh, and aims to solve the inverse population of the original population to expand the search space, perform one-time comparison between the original population and the inverse population, and screen out individuals with higher fitness values to form a new initial population. The inverse position of the population position generated by Tent mapping can be obtained by the following formula (16):
(2) Local optimal trap prevention strategy based on Cauchy variation disturbance
Aiming at the local optimal trap risk, the invention adopts a variation perturbation strategy, and the mayfly population with a certain proportion is selected at the end of the MA algorithm to perform position variation again, and the population is randomly hopped to a new area far away from the current position to detect whether a better position solution exists. The processing mode not only keeps the local searching capability of the original breeding population, but also increases the new global searching capability, and can reduce the probability of trapping in the local optimal trap. Cauchy variation is one of the most common variation disturbance methods, and is widely applied to various meta-heuristic algorithms, wherein a standard Cauchy distribution formula is shown as a formula (17). The invention randomly selects N from the offspring population of mayflies propagation mu t mayflies whose positions are mutated by the Cauchi mutation method as shown in formula (18). Combining the varied population and the original progeny population to form a new progeny population, sorting the populations according to the fitness value in order to keep the population quantity unchanged, and screening out the top N with excellent performance off Mayflies enter the next link of the algorithm as the final offspring population.
In the above formula (18), C i Is a random number generated according to equation (17).
MMA-ELM motion error prediction model
Combining the ELM network structure designed above and the improved MA optimization algorithm, the pseudo code of the final MMA-ELM motion error prediction model algorithm is shown in the following table:
compared with the existing robot motion error compensation algorithm, the method belongs to a data-based model-free algorithm, an accurate mathematical model is not required to be established, geometric errors and non-geometric errors can be simultaneously compensated, the method for compensating the robot motion error is provided, the motion error of the execution tail end of the robot is predicted based on an improved mayfly algorithm optimized extreme learning machine (MMA-ELM) network model, the algorithm model keeps the advantages of simple structure, high training speed and the like of an ELM network, and simultaneously overcomes the defects of unstable performance, difficulty in finding an optimal solution and the like of the algorithm model; therefore, the algorithm provided by the invention has the advantages of simple structure, high prediction precision, stable performance, high training speed and the like.
In order to verify the effectiveness of the motion error compensation algorithm provided by the invention, a motion error estimation experiment and a robot motion precision comparison experiment before and after motion error compensation are carried out on a 6-degree-of-freedom robot, and the experimental effects of various algorithms are contrastively analyzed; the experimental result shows that after the compensation is performed by the MMA-ELM algorithm, the absolute position error delta P of the executed tail end of the robot is reduced by 85.14% compared with that before the compensation, and it is obvious that the algorithm provided by the invention has the advantages of simple structure, high prediction precision, stable performance, high training speed and the like.
It should be noted that, in this document, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (7)
1. A robot motion error compensation method is characterized by comprising the following steps:
step one, training a robot motion error prediction model by using an acquired error data set;
and step two, correcting the robot motion planning output by using the trained motion error prediction model.
2. A robot motion error compensation method according to claim 1, wherein the error in step one isThe difference data set comprises a pose error data set
And driving joint target angular position data set
Where N is the number of samples contained in the data set.
3. A robot motion error compensation method according to claim 1, wherein the pose error data set
The obtaining method is as follows:
1. in the reachable working space of the execution tail end of the robot, N target point positions are randomly screened out to form a target pose data set
2. Controlling the execution tail end of the robot to reach the point positions, actually measuring the position and pose of each point position by using external measuring equipment, and forming an actually measured position and pose data set by the actually measured values
3. The difference between the actual measurement pose and the target pose of the robot execution end at these point locations is the motion error of the robot, which constitutes a pose error dataset
As a set of labels for training a motion error prediction model.
4. A robot motion error compensation method according to claim 1, wherein the driving joint targetAngular position data set
The obtaining method is as follows:
object pose data set
Input to the ideal inverse kinematics model f of the robot - (T ft ) In order to obtain
5. A method for compensating motion error of a robot according to claim 1, wherein the motion error prediction model in the first step is:
ELM network model, the training method of ELM network model adopts a novel metaheuristic optimization algorithm, namely the improved mayfly algorithm MMA;
the MMA-ELM model is abbreviated as the MMA-ELM model.
6. A robot motion error compensation method according to claim 5, characterized in that the improved mayfly algorithm MMA modifies the initialization strategy of the original mayfly algorithm using the improved Tent chaotic map and the reverse learning theory to improve the initial conditions and convergence speed of the mayfly algorithm;
the improved Tent chaotic mapping expression is as follows:
in the formula, z i Representing the value of the chaotic variable, a 4 E (0, 1) is a constant, and since the values generated by Tent mapping are not necessarily within the limited range of mayflies' population positions, it is also necessary to put their carriers into a set position solution space, as shown in the following equation:
x mi ＝x min +(x max -x min )z i ；
the inverse position of the population position generated by Tent mapping can be obtained by the following formula:
meanwhile, the Cauchy variation theory is adopted to improve the position updating strategy of the mayfly algorithm so as to reduce the probability of trapping the algorithm in the local optimal trap.
7. A robot motion error compensation method according to claim 5, wherein the specific process of the second step is:
1. planning a new robot execution end target point position in the working space, wherein the position and posture matrix of the point position is T ft It is input into an inverse kinematics model f - (T ft ) Then the target angle position theta of the robot driving joint can be obtained t ；
Theta is a function of t The method is not directly used for driving the robot, but the robot execution end pose error delta T under the instruction is estimated through a trained ELM network f ；
2. The original target pose T ft And the estimated pose error delta T f Adding to obtain a new target pose T fc ＝T ft +ΔT f ；
3. Will T fc Input into an inverse kinematics model f - (T fc ) The compensated angle position theta of the driving joint can be obtained c ，θ c Is a final angle command value for driving the robot.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211382761.9A CN115700414A (en) | 2022-11-07 | 2022-11-07 | Robot motion error compensation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211382761.9A CN115700414A (en) | 2022-11-07 | 2022-11-07 | Robot motion error compensation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115700414A true CN115700414A (en) | 2023-02-07 |
Family
ID=85121101
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211382761.9A Pending CN115700414A (en) | 2022-11-07 | 2022-11-07 | Robot motion error compensation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115700414A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117697769A (en) * | 2024-02-06 | 2024-03-15 | 成都威世通智能科技有限公司 | Robot control system and method based on deep learning |
-
2022
- 2022-11-07 CN CN202211382761.9A patent/CN115700414A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117697769A (en) * | 2024-02-06 | 2024-03-15 | 成都威世通智能科技有限公司 | Robot control system and method based on deep learning |
| CN117697769B (en) * | 2024-02-06 | 2024-04-30 | 成都威世通智能科技有限公司 | Robot control system and method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108803349B (en) | Optimal consistency control method and system for nonlinear multi-agent system | |
| Mishra et al. | Assembly sequence optimization using a flower pollination algorithm-based approach | |
| CN110900598B (en) | Robot three-dimensional motion space action simulation learning method and system | |
| CN109241291A (en) | Knowledge mapping optimal path inquiry system and method based on deeply study | |
| CN112484732B (en) | IB-ABC algorithm-based unmanned aerial vehicle flight path planning method | |
| CN108549210A (en) | Multiple no-manned plane based on BP neural network PID control cooperates with flying method | |
| CN111415051A (en) | Power material demand plan prediction method based on BP neural network | |
| CN114154401B (en) | Soil erosion modulus calculation method and system based on machine learning and observation data | |
| CN112859855A (en) | Robot multi-target path planning based on locust optimization algorithm | |
| CN115700414A (en) | Robot motion error compensation method | |
| CN113141012A (en) | Power grid power flow regulation and control decision reasoning method based on deep deterministic strategy gradient network | |
| Warsito et al. | Particle swarm optimization versus gradient based methods in optimizing neural network | |
| CN108955689A (en) | It is looked for food the RBPF-SLAM method of optimization algorithm based on adaptive bacterium | |
| CN115438842A (en) | Load prediction method based on adaptive improved dayflies and BP neural network | |
| Hasan et al. | Neural networks’ based inverse kinematics solution for serial robot manipulators passing through singularities | |
| Lan et al. | Cooperative Guidance of Multiple Missiles: A Hybrid Coevolutionary Approach | |
| Huang et al. | Automatic generation and optimization of stable assembly sequence based on ACO algorithm | |
| CN118428441A (en) | PID control algorithm based on improved enhanced dung beetle algorithm and counter-propagating neural network mixed optimization | |
| CN117669359A (en) | Ultrasonic welding process parameter recommendation method and system for connecting sheet | |
| Palangpour et al. | Recurrent neural network based predictions of elephant migration in a South African game reserve | |
| Warsito et al. | Particle swarm optimization to obtain weights in neural network | |
| Mairaj et al. | Identifying the optimal solutions of Bohachevsky test function using swarming algorithms | |
| Siebel et al. | Efficient learning of neural networks with evolutionary algorithms | |
| Kinoshita et al. | Solution of inverse kinematics by PSO based on split and merge of particles | |
| Tao et al. | Risk-aware multi-robot collaboration with arrival deadlines in uncertain environments |
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 |