Complex dynamic trajectory tracking method, device, robotic arm and storage medium based on incremental learning

Abstract

The present invention provides a complex dynamic trajectory tracking method, device, robotic arm, and storage medium based on incremental learning, relating to the field of control. The present invention utilizes an interval type-II evolutionary fuzzy neural network to obtain torque compensation data at the current moment, and then employs an incremental learning mechanism: the torque compensation data at the current moment is added to the torque control data at the previous moment to obtain the torque control data at the current moment; the torque control data at the current moment is then input into a tracking control model to obtain the joint drive data at the current moment. The obtained joint drive data is then used to drive the movement of each rotary joint, allowing the end effector of the robotic arm to track the target trajectory. This can overcome the cumulative tracking errors and unexpected dynamic changes in tracking complex dynamic target trajectories, ensuring high-precision tracking control of complex trajectories.

Description

Technical Field

[0001] The present invention relates to the field of control, and in particular to a complex dynamic trajectory tracking method, device, robotic arm and storage medium based on incremental learning. Background Art

[0002] With the rapid development of automation technology, industrial autonomous robots are increasingly being used in various fields. This has led to higher demands on the performance characteristics of industrial autonomous robots, particularly in terms of high-precision tracking control algorithms. Currently, a key challenge in robotics applications is designing control algorithms that can precisely guide end-effectors along user-defined or desired trajectories to meet the demands of high-end industrial applications such as aerospace assembly, laser cutting, and wing painting.

[0003] Some current robotic arm control algorithms are only capable of simple sinusoidal or cosine trajectory tracking. As the robot's operating environment becomes increasingly complex, the trajectories that the robotic arm's end effector must track are also becoming increasingly complex. Due to the inherent complexity and unpredictability of complex trajectories, existing robotic arm control algorithms often perform poorly, limiting the application of robotics in high-end applications.

[0004] Therefore, how to achieve high-precision tracking control of complex trajectories for industrial robotic arms is an urgent problem that needs to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide a complex dynamic trajectory tracking method, device, robotic arm and storage medium based on incremental learning to improve the problems existing in the prior art.

[0006] The embodiments of the present invention can be implemented as follows:

[0007] In a first aspect, the present invention provides a complex dynamic trajectory tracking method based on incremental learning, which is applied to a controller of a robotic arm, comprising:

[0008] Get the joint drive data and torque control data of the previous moment;

[0009] Determine the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target trajectory that is closest to the tracking point at the current moment;

[0010] Add the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment;

[0011] Input the torque control data at the current moment into the pre-trained tracking control model to obtain the joint drive data at the current moment;

[0012] The joint driving data at the current moment is used to drive the various rotation joints of the robotic arm to move, so that the end effector of the robotic arm tracks the target trajectory.

[0013] In a second aspect, the present invention provides a complex dynamic trajectory tracking device based on incremental learning, which is applied to a controller of a robotic arm, and the device comprises:

[0014] The acquisition module is used to obtain the joint drive data and torque control data of the previous moment;

[0015] A compensation calculation module is used to determine the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target track that is closest to the tracking point at the current moment;

[0016] The compensation calculation module is further configured to add the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment;

[0017] The drive calculation module is used to input the torque control data at the current moment into the pre-trained tracking control model to obtain the joint drive data at the current moment;

[0018] The drive control module is used to drive the various rotation joints of the robotic arm to move using the joint drive data at the current moment, so that the end effector of the robotic arm tracks the target trajectory.

[0019] In a third aspect, the present invention provides a robotic arm, comprising a controller, wherein the controller is used to implement a complex dynamic trajectory tracking method based on incremental learning as described in any one of the aforementioned embodiments.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the complex dynamic trajectory tracking method based on incremental learning described in any one of the aforementioned embodiments.

[0021] Compared with the prior art, an embodiment of the present invention provides a complex dynamic trajectory tracking method, device, robotic arm and storage medium based on incremental learning. At the current moment of tracking the target trajectory, the method is: obtaining the joint drive data and torque compensation data of the previous moment; determining the torque compensation data of the current moment based on the joint drive data of the previous moment and the trajectory point on the target trajectory closest to the tracking point at the current moment; adding the torque compensation data of the current moment to the torque control data of the previous moment to obtain the torque control data of the current moment; inputting the torque control data of the current moment into a pre-trained tracking control model to obtain the joint drive data of the current moment; and using the joint drive data of the current moment to drive the movement of each rotation joint of the robotic arm so that the end effector of the robotic arm tracks the target trajectory. The present invention adds the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment; then the torque control data at the current moment is input into the tracking control model to obtain the joint drive data at the current moment. In this way, the obtained joint drive data is used to drive the movement of each rotary joint, which can overcome the cumulative tracking error and unexpected dynamic changes in tracking complex dynamic target trajectories, and ensure high-precision tracking control of complex trajectories. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A flowchart of a complex dynamic trajectory tracking method based on incremental learning is provided in an embodiment of the present invention.

[0024] Figure 2 A schematic structural diagram of a TS fuzzy system provided by an embodiment of the present invention.

[0025] Figure 3 An overall control block diagram provided by an embodiment of the present invention.

[0026] Figure 4 A schematic diagram of a robotic arm provided in an embodiment of the present invention.

[0027] Figure 5 A schematic structural diagram of a complex dynamic trajectory tracking device based on incremental learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0029] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0030] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0031] In addition, the terms "first", "second", etc., if used, are merely used to distinguish and describe, and should not be understood as indicating or implying relative importance.

[0032] It should be noted that, in the absence of conflict, the features in the embodiments of the present invention may be combined with each other.

[0033] Please refer to Figure 1 , Figure 1 A flowchart of a complex dynamic trajectory tracking method based on incremental learning provided by an embodiment of the present invention is provided. The execution subject of the method may be a controller of a robotic arm. The method includes the following steps S101 to S100:

[0034] S101: Acquire joint drive data and torque compensation data at the previous moment.

[0035] The robotic arm may include a base, a plurality of rotational joints connected by a plurality of connecting rods, and an end effector. The end effector of the robotic arm in the present invention may track a target trajectory.

[0036] In this embodiment, the joint drive data at the last moment may include the joint angle, joint angular velocity, and joint angular acceleration of each rotary joint at the last moment. The torque control data at the last moment may include the torque control signal of each rotary joint at the last moment.

[0037] S102 : Determine the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target track that is closest to the tracking point at the current moment.

[0038] In this embodiment, the torque compensation data at the current moment may include a torque compensation signal of each rotary joint at the current moment.

[0039] S104 : Add the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment.

[0040] In this embodiment, the torque control data at the current moment may include the torque control signal of each rotary joint at the current moment. The torque compensation signal of each rotary joint at the current moment is added to the torque control signal of each rotary joint at the previous moment to obtain the torque control signal of each rotary joint at the current moment.

[0041] S105 , inputting the torque control data at the current moment into a pre-trained tracking control model to obtain the joint drive data at the current moment.

[0042] In this embodiment, the joint driving data at the current moment may include the joint angle, joint angular velocity, and joint angular acceleration of each rotary joint of the robotic arm at the current moment.

[0043] S106 : Using the joint driving data at the current moment, drive each rotation joint of the robotic arm to move, so that the end effector of the robotic arm tracks the target trajectory.

[0044] The complex dynamic trajectory tracking method based on incremental learning provided by the present invention adopts an incremental learning mechanism after obtaining the torque compensation data at the current moment: the torque control data at the previous moment is compensated by the torque compensation data at the current moment, thereby obtaining the torque control data at the current moment and inputting it into the tracking control model. The joint drive data at the current moment obtained in this way drives the movement of each rotational joint to enable the end effector of the robotic arm to track the target trajectory, which can effectively overcome the cumulative tracking errors and unexpected dynamic changes in tracking the complex dynamic target trajectory, and ensure high-precision tracking control of the complex trajectory.

[0045] Optionally, the tracking control model is obtained based on the TS fuzzy model (also called the inverse model). Here, the training process of the tracking control model is first introduced.

[0046] First, the obtained dataset can be expressed as:

[0047]

[0048] In formula (1), the dataset {x} N It includes N data samples, among which the nth data sample x n The dimension is C, n = 1, 2, ..., N; the data sample x nIncluding the torque control signal of each rotation joint, joint angle, joint angular velocity and joint angular acceleration.

[0049] A data sample can be collected by applying a randomly generated torque control signal to each rotating joint of the robotic arm, and then collecting the joint angle, joint angular velocity, and joint angular acceleration of each rotating joint. By continuously applying the randomly generated torque control signal to each rotating joint of the robotic arm, multiple data samples can be collected.

[0050] The dataset {x} N Input into the pre-built TS fuzzy system, so that the TS fuzzy system can be based on the data set {x} N The mapping relationship between the torque control signal of each rotation joint and the shutdown angle information is learned from each data sample in the .

[0051] The TS fuzzy system uses the AnYa fuzzy system. In the AnYa fuzzy system, for an input data sample x n , can be expressed by AnYa type fuzzy rules as follows:

[0052] Rule i′: If (x n ~Ξ i′ ), then (p i′ ) (2)

[0053] In formula (2), i′=1,2,...,R′, R′ represents the number of linear models in the modeling process of an AnYa fuzzy system (i.e., the number of AnYa fuzzy systems), i′ represents the antecedent part of the i′th linear model; p i′ Represents the consequent part of the i′th linear model.

[0054] According to the first-order polynomial input in the posterior of the i′th linear model, the posterior part in equation (2) is expressed as:

[0055] p i′ (k) = z′ 0i′ (k)+z′ 1i′ (k)x1(k)+…+z′ ni′ (k)x n (k) (3)

[0056] In formula (2), k represents the current time, p i′ (k) represents the subsequent output, z′ 0i′ (k),z′ 1i′ (k),…,z′ ni′ (k) is the consequent parameter sequence of the i′th linear model.

[0057] When the consequent output of each linear model is multiplied by the normalized value of its activation strength, the final output of an AnYa fuzzy system can be expressed as:

[0058]

[0059] In formula (4), p is the total output of AnYa fuzzy system, represents the activation degree of the i′th linear model in an AnYa fuzzy system, p i′ It represents the consequent output of the i′th linear model in an AnYa fuzzy system.

[0060] The modeling structure of TS fuzzy system is as follows Figure 2 As shown in Figure 1, the TS fuzzy system is a collection of many sub-fuzzy systems consisting of an input layer, a set of hidden layers and an output layer. The hidden layer of the TS fuzzy system consists of multiple AnYa type fuzzy systems connected to each other.

[0061] A data sample x n State variables in Including the joint angle, joint angular velocity, and joint angular acceleration of each rotational joint. Figure 2 , state variables Scan the input layer through the sliding window and obtain the intermediate output variables through parallel calculation Intermediate output variables is fed into the first hidden layer to produce the intermediate output variable The functions of the following hidden layers are similar and will not be described in detail here. Finally, the output of the last hidden layer is Enter the output layer to produce the total output of the TS fuzzy system

[0062] According to the above analysis, Figure 2 The input-output relationship of the lth layer in can be described as:

[0063]

[0064] In formula (5), and Represent the input and output of layer l respectively.

[0065] The input of layer l as follows

[0066]

[0067] In formula (6), L is the number of AnYa fuzzy systems used in the l-1th layer; p L represents the total output of the Lth AnYa fuzzy system in the l-1th layer, p LFor the calculation of , see the above formula (4).

[0068] Assume that the output of layer l-1 is used as the input of layer l (denoted as ), then the mathematical modeling of the entire TS fuzzy system can be expressed as the following composite function:

[0069]

[0070] In formula (7), Represents the torque control signal of each rotary joint of the robot arm, represents the joint angle, joint angular velocity, and joint angular acceleration of each rotary joint of the manipulator. Therefore, Equation (7) as a whole reflects the actual mapping relationship between the torque control signal of each rotary joint of the manipulator and the joint angle, angular velocity, and angular acceleration.

[0071] It can be understood that since the TS fuzzy model is also called the inverse model, the input in the training phase is The essence is the joint angle, joint angular velocity, and joint angular acceleration of each rotating joint, and the output The input of the tracking control model obtained after training is the torque control data (i.e., the torque control signal of each rotating joint), and the output is the joint drive data (i.e., the joint angle, joint angular velocity, and joint angular acceleration of each rotating joint).

[0072] Optionally, the controller runs an interval-type-two evolved fuzzy neural network (IT2EFNN), which generates torque compensation data for error compensation. The IT2EFNN initializes its structure from the initial moment of target trajectory tracking. During this initialization process, a single fuzzy model is used. The IT2EFNN updates each moment of the end effector tracking the target trajectory and optimizes by eliminating redundant rules, thereby improving tracking control accuracy and system responsiveness.

[0073] The following introduces the process of determining the torque compensation data at the current moment in combination with the interval type-2 evolving fuzzy neural network.

[0074] Combine Figure 3 In the above step S102, the process of "determining the torque compensation data at the current moment based on the joint drive data at the previous moment and the trajectory point on the target trajectory closest to the tracking point at the current moment" can include the following sub-steps S1021 to S1026.

[0075] The step of determining the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target track closest to the tracking point at the current moment includes:

[0076] S1021. Determine the tracking point of the end effector at the current moment based on the joint drive data at the previous moment.

[0077] In this embodiment, combined with Figure 3 , at the last moment, using the joint drive data of the last moment The various rotation joints of the robotic arm can be driven to move, so that the end effector of the robotic arm tracks the target trajectory, thereby obtaining the tracking point r of the end effector at the current moment. d' (k).

[0078] For an optional example, see Figure 4 , taking the two rotating joints (joint 1 and joint 2) of the robot arm as an example, the joint drive data at the previous moment includes the joint angle q1 of joint 1 and the joint angle q2 of joint 2, then the tracking point r of the end effector at the current moment d' The coordinates of (k) can be converted using the following formula:

[0079]

[0080] In formula (8), (x d' ,y d' ) represents the tracking point r at the current moment d' (k), l1 and l2 are the lengths of link 1 corresponding to joint 1 and link 2 corresponding to joint 2, respectively.

[0081] Figure 4 This example is merely an example, and the present invention does not limit the structure of the robotic arm or the number of rotation joints of the robotic arm.

[0082] S1022: Determine the track point closest to the current tracking point from the target track.

[0083] In this embodiment, combined with Figure 2 , the tracking point r on the target trajectory from the current moment d' (k) The nearest trajectory point is r d (k).

[0084] S1023: Determine the tracking error at the current moment based on the trajectory point and tracking point at the current moment.

[0085] S1024: Obtain the tracking error at the previous moment, and subtract the tracking error at the current moment from the tracking error at the previous moment to obtain the error change rate at the current moment.

[0086] In this embodiment, combined with Figure 2 , the tracking error and error change rate at the current moment are as follows:

[0087]

[0088] In formula (9), k represents the current time, ξ(k) represents the tracking error at the kth time, Δξ(k) represents the error change rate at the kth time, and ξ(k-1) represents the tracking error at the k-1th time.

[0089] S1025. Combine the error change rate and the tracking error at the current moment to obtain an error combination vector, and expand the error combination vector to obtain the error vector at the current moment.

[0090] In this embodiment, the error vector at the current moment is: in, is the error combination vector.

[0091] S1026. Input the error vector at the current moment into the interval type-II evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment.

[0092] In this embodiment, the interval-type-2 evolutionary fuzzy neural network may include multiple fuzzy models and a Gaussian membership function and consequent parameter sequence of each fuzzy model.

[0093] Therefore, the process of "inputting the error vector at the current moment into the interval type-2 evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment" in step S1026 may include the following sub-steps S10261 to S10263:

[0094] S10261. Based on the Gaussian membership function of each fuzzy model at the current moment, calculate the lower activation strength and the upper activation strength of the error vector at the current moment for each fuzzy model.

[0095] In this embodiment, for interval type-2 fuzzy sets, the Gaussian membership function can include a lower membership function and an upper membership function. Therefore, based on the Gaussian membership function of each fuzzy model at the current moment, the formula for calculating the lower activation strength and upper activation strength of the error vector at the current moment for each fuzzy model is as follows:

[0096]

[0097]

[0098] In formula (10), k represents the current time, are the lower activation intensity and upper activation intensity of the i-th fuzzy model respectively; Represents the error vector x at the kth moment k The j-th vector value in , i=1,2,…,R, R represents the number of fuzzy models;

[0099] In formula (11), is the lower membership function of the i-th fuzzy model, is the upper membership function of the i-th fuzzy model; express Obey As the center point, the Gaussian distribution with σ as the variance, express Obey is a Gaussian distribution with σ as the center and σ as the variance.

[0100] S10262. Calculate the weight coefficient of each fuzzy model at the current moment based on the lower activation strength and upper activation strength of each fuzzy model according to the error vector at the current moment.

[0101] In this embodiment, based on the lower activation strength and upper activation strength of each fuzzy model by the error vector at the current moment, the formula for calculating the weight coefficient of each fuzzy model at the current moment is as follows:

[0102]

[0103] In formula (12), W i,k is the weight coefficient of the i-th fuzzy model at the k-th moment, reflecting the activation level of the i-th fuzzy model at the k-th moment; α and β are both weighting coefficients, satisfying α+β=1.

[0104] S10263. Calculate the torque compensation data at the current moment based on the weight coefficient of each fuzzy model at the current moment, the error vector at the current moment, and the consequent parameter sequence of each fuzzy model at the current moment.

[0105] In this embodiment, based on the weight coefficient of each fuzzy model at the current moment, the error vector at the current moment, and the consequent parameter sequence of each fuzzy model at the current moment, the formula for calculating the torque compensation data at the current moment is as follows:

[0106]

[0107] In formula (13), is the torque compensation data at the kth moment; Φ i,k is the consequent parameter sequence of the i-th fuzzy model at the k-th moment;

[0108] In formula (13), W(k)=[W 1,k ,W 2,k …,W R,k ], W(k) is the weight coefficient vector of all fuzzy models;

[0109] In formula (13), S(k) is the coefficient matrix of all fuzzy models, which can be expressed as:

[0110]

[0111] In formula (14),

[0112] The above steps S1021 to S1026 are the complete process of determining the torque compensation data at the current moment.

[0113] In an optional implementation, after obtaining the torque compensation data at the current moment, the method may further include:

[0114] S103 , updating the interval type-2 evolved fuzzy neural network to obtain the interval type-2 evolved fuzzy neural network at the next moment.

[0115] In this embodiment, the updating process of the interval type-2 evolving fuzzy neural network includes adding and optimizing the fuzzy model. Therefore, the sub-steps of step S103 may include S1031 to S1032.

[0116] S1031. Based on the error vector at the current moment, the interval type-2 evolved fuzzy neural network is updated to obtain an updated interval type-2 evolved fuzzy neural network.

[0117] In this embodiment, combined with Figure 2 Based on the error vector at the current moment, the interval type-2 evolving fuzzy neural network can be updated based on a learning algorithm. The learning algorithm can include the following steps S10311 to S1031C.

[0118] S10311. Obtain the error vector of the k moments obtained by tracking the target trajectory, the center point potential energy at the k-1th moment, the cluster center potential energy at the first moment, the center point potential energy of each fuzzy model at the k-1th moment, the weight coefficient of each fuzzy model at the kth moment, and the intermediate parameter matrix of each fuzzy model at the kth moment; wherein the kth moment is the current moment.

[0119] S10312. Based on the error vectors at k moments, calculate the fusion term at the kth moment, the fusion term at the previous k-1 moments, and the cross-fusion term at k moments.

[0120] The calculation formulas for the fusion term at the kth moment, the fusion term at the previous k-1 moments, and the cross-fusion term at the kth moment are as follows:

[0121]

[0122] In formula (15), χk represents the fusion term at the kth moment; Represents the error vector x at the kth moment k The j-th vector value in , λ k Represents the fusion term of the first k-1 moments; is the error vector x at the lth moment l The j-th vector value in , ζ k Represents the cross-fusion term of k moments.

[0123] S10313. Calculate the potential energy of the error vector at the kth moment based on the fusion term at the kth moment, the fusion terms at the previous k-1 moments, and the cross-fusion terms at the kth moment.

[0124] In this embodiment, based on the fusion term at the kth moment, the fusion term at the previous k-1 moments, and the cross-fusion term at the kth moment, the formula for calculating the potential energy of the error vector at the kth moment is as follows:

[0125]

[0126] In formula (16), is the error vector x at the kth moment k potential energy.

[0127] S10314. Based on the error vector at the kth moment, the error vector at the k-1th moment, and the center point potential energy of each fuzzy model at the k-1th moment, calculate the center point potential energy of each fuzzy model at the kth moment.

[0128] In this embodiment, based on the error vector at the kth moment, the error vector at the k-1th moment, and the center point potential energy of each fuzzy model at the k-1th moment, the formula for calculating the center point potential energy of each fuzzy model at the kth moment is as follows:

[0129]

[0130] In formula (17), is the central error vector corresponding to the construction of the i-th fuzzy model, Represents the potential energy of the center point of the i-th fuzzy model at the k-th moment; Represents the potential energy of the center point of the i-th fuzzy model at the k-1th moment; Represents the error vector x at the kth moment k The jth vector value in and the error vector x at the k-1th moment k-1 The distance between the j-th vector values ​​in .

[0131] S10315. Based on the central point potential energy of all fuzzy models at the kth moment, determine the cluster center potential energy at the kth moment corresponding to the central error vector at the kth moment.

[0132] In this embodiment, the cluster center potential energy at the kth moment is: represents the cluster center potential energy at the kth moment; R represents the number of fuzzy models; := is the initialization symbol, which is used to assign values; is the central error vector at the kth moment,

[0133] S10316. Based on the cluster center potential energy at the k-th moment, reinitialize the center point potential energy of each fuzzy model at the k-th moment.

[0134] In this embodiment, since the center error vector at the kth moment And for Except Other central error vectors According to other central error vectors arrive The inverse relationship between the distance and other center error vectors The corresponding center point potential energy is relative to the cluster center potential energy at the kth moment Based on this, the formula for reinitializing the potential energy of the central point of each fuzzy model at the kth moment is as follows:

[0135]

[0136] In formula (18), e is a natural constant, ι is a positive constant. :=left side It is the potential energy of the center point of the i-th fuzzy model at the k-th moment after reassignment, and can subsequently participate in the calculation of the potential energy of the center point of the i-th fuzzy model at the k+1-th moment.

[0137] S10317. Calculate the distance between the error vector at the kth moment and the new center point potential of each fuzzy model at the kth moment, and determine the shortest distance from the multiple distances obtained by calculation.

[0138] S10318. Construct a first update condition based on a preset potential energy upper bound, and construct a second update condition based on the preset potential energy upper bound, the preset potential energy lower bound, the cluster center potential energy at the first moment, and the shortest distance.

[0139] In this embodiment, the first update condition is formula (19), and the second update condition is formula (20):

[0140]

[0141]

[0142] in, To preset the upper bound of potential energy, To preset the lower bound of potential energy, is the cluster center potential energy at the first moment, δ min is the shortest distance, and r is a constant.

[0143] S10319. If the potential energy of the error vector at the kth moment satisfies the first update condition, or the potential energy of the error vector at the kth moment and the potential energy of the cluster center at the kth moment satisfy the second update condition, then based on the central error vector at the kth moment, determine the first center point and the second center point, and based on the first center point and the second center point, construct the Gaussian membership function of the new fuzzy model at the k+1th moment.

[0144] In this embodiment, if the potential energy of the error vector at the kth moment is If the first update condition in the above formula (19) is satisfied, the first center point and the second center point are determined by the following formula (21). Alternatively, if the potential energy of the error vector at the kth moment is satisfy And the cluster center potential energy at the kth moment satisfy The first center point and the second center point are also determined by the following formula (21).

[0145] The calculation method of the first center point and the second center point is:

[0146]

[0147] In formula (21), the new fuzzy model is the R+1th fuzzy model, is the first center point, is the second center point, and Δ is the fixed width.

[0148] Among them, the formula of the Gaussian membership function of the R+1th fuzzy model at the k+1th moment is:

[0149]

[0150] In formula (22), Represents the error vector x at the k+1th moment k+1 The j-th vector value in ; are the upper membership function and lower membership function of the R+1th fuzzy model respectively; express Obey As the center point, the Gaussian distribution with σ as the variance; express Obey is a Gaussian distribution with σ as the center and σ as the variance.

[0151] S1031A: Calculate the optimal control torque rate at the kth moment based on the error vector at the kth moment and the weight coefficient of each fuzzy model at the kth moment.

[0152] In this embodiment, based on the error vector at the kth moment and the weight coefficient of each fuzzy model at the kth moment, the formula for calculating the optimal control torque rate at the kth moment is as follows:

[0153]

[0154] In formula (23), is the optimal control torque rate at the kth moment; W i,k is the weight coefficient of the i-th fuzzy model at the k-th moment, is the ideal optimal parameter of the i-th fuzzy model, ν i is the intrinsic approximation error of the i-th fuzzy model.

[0155] S1031B, based on the error vector at the kth moment and the intermediate parameter matrix and consequent parameter sequence of each fuzzy model at the kth moment, respectively determine the intermediate parameter matrix of each fuzzy model at the k+1th moment.

[0156] In this embodiment, the intermediate parameter matrix of each fuzzy model at the k+1th time is used to update the consequent parameter sequence of each fuzzy model at the k+1th time. The formula for calculating the intermediate parameter matrix of the i-th fuzzy model at the k+1th time is as follows:

[0157]

[0158] In formula (24), Ω i,k+1 The intermediate parameter matrix of the i-th fuzzy model at the k+1-th moment; Ω i,k is the intermediate parameter matrix of the i-th fuzzy model at the k-th moment, Φ i,k is the consequent parameter sequence of the i-th fuzzy model at the k-th moment, i = 1, 2, …, R + 1.

[0159] S1031C, based on the error vector and optimal control torque rate at the kth moment and the intermediate parameter matrix and consequent parameter sequence of each fuzzy model at the kth moment, respectively determine the consequent parameter sequence of each fuzzy model at the k+1th moment.

[0160] In this embodiment, the formula for calculating the consequent parameter sequence of the i-th fuzzy model at the k+1-th time is as follows:

[0161]

[0162] In formula (25), Φ i,k+1 is the consequent parameter sequence of the i-th fuzzy model at the k+1-th moment.

[0163] The above steps S10311 to S1031C are the process of adding a new fuzzy model to the interval type-2 evolved fuzzy neural network during updating to obtain an updated interval type-2 evolved fuzzy neural network.

[0164] S1032. Calculate the utility value of each fuzzy model in the updated interval type-2 evolving fuzzy neural network, and remove the fuzzy models whose utility values ​​are lower than a preset utility threshold and their Gaussian membership functions and consequent parameter sequences to obtain the interval type-2 evolving fuzzy neural network at the next moment.

[0165] In this embodiment, the interval type-2 evolutionary fuzzy neural network at the next moment is used to obtain the torque compensation data at the next moment. In the updated interval type-2 evolutionary fuzzy neural network, the formula for calculating the utility value of each fuzzy model is as follows:

[0166]

[0167] In formula (26), π k,i Represents the utility value of the i-th fuzzy model at the k-th moment. i represents the creation time of the i-th fuzzy model, W l,i is the weight coefficient of the i-th fuzzy model at the l-th moment, reflecting the activation level of the i-th fuzzy model at the l-th moment.

[0168] If the utility value of the i-th fuzzy model at the k-th moment is π k,i Satisfy: π k,i <π0, then it means that the i-th fuzzy model is a redundant rule and it is necessary to delete the i-th fuzzy model and its Gaussian membership function and consequent parameter sequence. Where π0 is the preset utility threshold, which can be set to 0.1.

[0169] It should be noted that the execution order of the steps in the above method embodiment is not limited by the step numbers, and the execution order of the steps shall be based on the actual application situation.

[0170] In order to execute the corresponding steps in the above method embodiment and various possible implementation methods, an implementation of a complex dynamic trajectory tracking device based on incremental learning is given below.

[0171] See Figure 5 , Figure 5 The schematic diagram of the structure of the complex dynamic trajectory tracking device based on incremental learning provided by an embodiment of the present invention is shown. The complex dynamic trajectory tracking device based on incremental learning 200 is applied to the controller of the robotic arm, and the device includes:

[0172] An acquisition module 210 is used to acquire joint drive data and torque compensation data at the previous moment;

[0173] The compensation calculation module 220 is used to determine the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target track that is closest to the tracking point at the current moment;

[0174] The compensation calculation module 220 is further configured to add the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment;

[0175] The drive calculation module 230 is used to input the torque control data at the current moment into the pre-trained tracking control model to obtain the joint drive data at the current moment;

[0176] The drive control module 240 is used to drive the various rotation joints of the robotic arm to move using the joint drive data at the current moment, so that the end effector of the robotic arm tracks the target trajectory.

[0177] Optionally, the controller runs an interval type-2 evolutionary fuzzy neural network, which is updated at each moment when the end effector tracks the target trajectory. The compensation calculation module 220, in determining the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target trajectory closest to the tracking point at the current moment, can specifically be configured to: determine the tracking point of the end effector at the current moment based on the joint drive data at the previous moment; determine the track point closest to the tracking point at the current moment from the target trajectory; determine the tracking error at the current moment based on the track point and the tracking point; obtain the tracking error at the previous moment, and subtract the tracking error at the current moment from the tracking error at the previous moment to obtain the error change rate at the current moment; combine the error change rate with the tracking error at the current moment to obtain an error combination vector, and expand the error combination vector to obtain the error vector at the current moment; and input the error vector at the current moment into the interval type-2 evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment.

[0178] Optionally, the interval type-2 evolutionary fuzzy neural network includes multiple fuzzy models and a Gaussian membership function and consequent parameter sequence of each fuzzy model. The compensation calculation module 220, in the process of inputting the error vector at the current moment into the interval type-2 evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment, can be specifically used to: calculate the lower activation intensity and upper activation intensity of the error vector at the current moment for each fuzzy model based on the Gaussian membership function of each fuzzy model at the current moment; calculate the weight coefficient of each fuzzy model at the current moment based on the lower activation intensity and upper activation intensity of the error vector at the current moment for each fuzzy model; calculate the torque compensation data at the current moment based on the weight coefficient of each fuzzy model at the current moment, the error vector at the current moment, and the consequent parameter sequence of each fuzzy model at the current moment.

[0179] Optionally, the complex dynamic trajectory tracking device 200 based on incremental learning further includes an updating module 250. After obtaining the torque compensation data at the current moment, the updating module 250 is used to update the interval type-2 evolutionary fuzzy neural network. Specifically, the updating module 250 can be used to: update the interval type-2 evolutionary fuzzy neural network based on the error vector at the current moment to obtain an updated interval type-2 evolutionary fuzzy neural network; calculate the utility value of each fuzzy model in the updated interval type-2 evolutionary fuzzy neural network, and remove the fuzzy model whose utility value is lower than a preset utility threshold and its Gaussian membership function and consequent parameter sequence to obtain the interval type-2 evolutionary fuzzy neural network at the next moment; wherein, the interval type-2 evolutionary fuzzy neural network at the next moment is used to obtain the torque compensation data at the next moment.

[0180] Optionally, the update module 250 is used to update the interval type-2 evolutionary fuzzy neural network based on the error vector at the current moment, and in the process of obtaining the updated interval type-2 evolutionary fuzzy neural network, it can be specifically used to: obtain the error vector of k moments obtained by tracking the target trajectory, the center point potential energy of the k-1th moment, the cluster center potential energy of the first moment, the center point potential energy of each fuzzy model at the k-1th moment, the weight coefficient of each fuzzy model at the kth moment, and the intermediate parameter matrix of each fuzzy model at the kth moment; wherein the kth moment is the current moment; based on the error vector at the kth moment, calculate the fusion term at the kth moment; based on the previous k-1 The error vector at the kth moment is used to calculate the fusion term of the previous k-1 moments; based on the error vector at the kth moment, the cross-fusion term at the kth moment is calculated; based on the fusion term at the kth moment, the fusion term at the previous k-1 moments and the cross-fusion term at the kth moment, the potential energy of the error vector at the kth moment is calculated; based on the error vector at the kth moment, the error vector at the k-1th moment and the center point potential energy of each fuzzy model at the k-1th moment, the center point potential energy of each fuzzy model at the kth moment is calculated respectively; based on the center point potential energy of all the fuzzy models at the kth moment, the cluster center potential energy at the kth moment corresponding to the central error vector at the kth moment is determined; based on the kth moment Cluster center potential energy, reinitialize the center point potential energy of each fuzzy model at the kth moment; calculate the distance between the error vector at the kth moment and the new center point potential energy of each fuzzy model at the kth moment, and determine the shortest distance from the multiple distances calculated; construct a first update condition based on a preset potential energy upper bound, and construct a second update condition based on the preset potential energy upper bound, the preset potential energy lower bound, the cluster center potential energy at the first moment and the shortest distance; if the potential energy of the error vector at the kth moment meets the first update condition, or the potential energy of the error vector at the kth moment and the cluster center potential energy at the kth moment meet the second update condition, then based on the center point potential energy at the kth moment A central error vector is obtained, a first central point and a second central point are determined, and based on the first central point and the second central point, a Gaussian membership function of a new fuzzy model at the k+1th moment is constructed; based on the error vector at the kth moment and the weight coefficient of each of the fuzzy models at the kth moment, the optimal control torque rate at the kth moment is calculated; based on the error vector at the kth moment and the intermediate parameter matrix and consequent parameter sequence of each of the fuzzy models at the kth moment, the intermediate parameter matrix of each of the fuzzy models at the k+1th moment is determined respectively; wherein the intermediate parameter matrix of each of the fuzzy models at the k+1th moment is used to update the consequent parameter sequence of each of the fuzzy models at the k+1th moment;Based on the error vector and optimal control torque rate at the kth moment and the intermediate parameter matrix and consequent parameter sequence of each fuzzy model at the kth moment, respectively determine the consequent parameter sequence of each fuzzy model at the k+1th moment.

[0181] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the complex dynamic trajectory tracking device 200 based on incremental learning described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0182] An embodiment of the present invention further provides a robotic arm, comprising a controller for implementing the aforementioned complex dynamic trajectory tracking method based on incremental learning. The controller may be an integrated circuit chip having signal processing capabilities. The controller may be a general-purpose processor, including: a CPU (Central Processing Unit), an NP (Network Processor), an SoC (System on Chip), etc.; it may also be: a DSP (Digital Signal Processing), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0183] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer-readable storage medium implements the complex dynamic trajectory tracking method based on incremental learning disclosed in the above embodiment. The computer-readable storage medium may be, but is not limited to, RAM (Random Access Memory), ROM (Read Only Memory), FLASH (Flash Memory), PROM (Programmable Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electric Erasable Programmable Read-Only Memory), etc.

[0184] In summary, the embodiments of the present invention provide a complex dynamic trajectory tracking method, device, robotic arm and storage medium based on incremental learning. At the current moment of tracking the target trajectory, the method is: obtaining the joint drive data and torque compensation data of the previous moment; determining the torque compensation data of the current moment based on the joint drive data of the previous moment and the trajectory point on the target trajectory closest to the tracking point at the current moment; adding the torque compensation data of the current moment to the torque control data of the previous moment to obtain the torque control data of the current moment; inputting the torque control data of the current moment into a pre-trained tracking control model to obtain the joint drive data of the current moment; using the joint drive data at the current moment to drive the movement of each rotation joint of the robotic arm, so that the end effector of the robotic arm tracks the target trajectory. The present invention adds the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment; then the torque control data at the current moment is input into the tracking control model to obtain the joint drive data at the current moment. In this way, the obtained joint drive data is used to drive the movement of each rotary joint, which can overcome the cumulative tracking error and unexpected dynamic changes in tracking complex dynamic target trajectories, and ensure high-precision tracking control of complex trajectories.

[0185] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A complex dynamic trajectory tracking method based on incremental learning, characterized in that: Controllers for robotic arms include: Get the joint drive data and torque control data of the previous moment; Determine the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target trajectory that is closest to the tracking point at the current moment; Add the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment; Input the torque control data at the current moment into the pre-trained tracking control model to obtain the joint drive data at the current moment; Using the joint drive data at the current moment, driving each rotation joint of the robotic arm to move, so that the end effector of the robotic arm tracks the target trajectory; The controller runs an interval type-2 evolving fuzzy neural network, which is updated at each moment when the end effector tracks the target trajectory; the step of determining the torque compensation data at the current moment based on the joint drive data at the previous moment and the trajectory point on the target trajectory closest to the tracking point at the current moment includes: Determining a tracking point of the end effector at a current moment based on joint drive data at a previous moment; Determining the track point closest to the current tracking point from the target track; determining a tracking error at a current moment based on the trajectory point and the tracking point; Obtain the tracking error at the previous moment, and subtract the current tracking error from the previous moment to obtain the error change rate at the current moment; Combining the error change rate and the tracking error at the current moment to obtain an error combination vector, and expanding the error combination vector to obtain an error vector at the current moment; The error vector at the current moment is input into the interval type-II evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment.

2. The method according to claim 1, characterized in that The interval type-2 evolutionary fuzzy neural network includes a plurality of fuzzy models and a Gaussian membership function and consequent parameter sequence of each fuzzy model; The step of inputting the error vector at the current moment into the interval type-2 evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment includes: Calculating the lower activation strength and the upper activation strength of the error vector at the current moment for each of the fuzzy models based on the Gaussian membership function of each of the fuzzy models at the current moment; Calculating a weight coefficient of each fuzzy model at the current moment based on the lower activation strength and the upper activation strength of each fuzzy model according to the error vector at the current moment; The torque compensation data at the current moment is calculated based on the weight coefficient of each fuzzy model at the current moment, the error vector at the current moment, and the consequent parameter sequence of each fuzzy model at the current moment.

3. The method according to claim 2, characterized in that The Gaussian membership function includes a lower membership function and an upper membership function; Wherein, based on the Gaussian membership function of each fuzzy model at the current moment, the formula for calculating the lower activation strength and the upper activation strength of the error vector at the current moment for each fuzzy model is as follows: Where, Indicates the current moment, 、 Respectively The lower activation intensity and upper activation intensity of the fuzzy model; Representative The error vector at each moment The vector values, ; For the The lower membership function of the fuzzy model is For the The upper membership function of a fuzzy model; express Obey As the center point, is a Gaussian distribution with a variance of express Obey As the center point, is a Gaussian distribution with variance; Based on the lower activation strength and upper activation strength of each fuzzy model according to the error vector at the current moment, the formula for calculating the weight coefficient of each fuzzy model at the current moment is as follows: Where, For the The fuzzy model is The weight coefficient of each moment; 、 are weighted coefficients, satisfying ; is the number of the fuzzy models; Based on the weight coefficient of each fuzzy model at the current moment, the error vector at the current moment, and the consequent parameter sequence of each fuzzy model at the current moment, the formula for calculating the torque compensation data at the current moment is as follows: Where, For the Torque compensation data at each moment; For the The fuzzy model is The sequence of consequent parameters at each moment.

4. The method according to claim 1, wherein The interval type-2 evolutionary fuzzy neural network includes a plurality of fuzzy models and a Gaussian membership function and consequent parameter sequence of each fuzzy model; After obtaining the torque compensation data at the current moment, the method further includes: a step of updating the interval type-2 evolutionary fuzzy neural network, which includes: Based on the error vector at the current moment, the interval type-2 evolving fuzzy neural network is updated to obtain an updated interval type-2 evolving fuzzy neural network; The utility value of each fuzzy model in the updated interval type-2 evolving fuzzy neural network is calculated, and the fuzzy models whose utility values ​​are lower than a preset utility threshold and their Gaussian membership functions and consequent parameter sequences are removed to obtain the interval type-2 evolving fuzzy neural network at the next moment; wherein the interval type-2 evolving fuzzy neural network at the next moment is used to obtain the torque compensation data at the next moment.

5. The method according to claim 4, characterized in that The step of updating the interval type-2 evolved fuzzy neural network based on the error vector at the current moment to obtain an updated interval type-2 evolved fuzzy neural network includes: Obtaining the error vectors at k moments obtained by tracking the target trajectory, the center point potential energy at the k-1th moment, the cluster center potential energy at the first moment, the center point potential energy of each fuzzy model at the k-1th moment, the weight coefficient of each fuzzy model at the kth moment, and the intermediate parameter matrix of each fuzzy model at the kth moment; wherein the kth moment is the current moment; Based on the error vector at the kth moment, calculate the fusion term at the kth moment; Based on the error vector of the previous k-1 moments, calculate the fusion term of the previous k-1 moments; Based on the error vectors at k moments, calculate the cross-fusion terms at k moments; Calculate the potential energy of the error vector at the kth moment based on the fusion term at the kth moment, the fusion term at the previous k-1 moments, and the cross-fusion term at the kth moment; Calculating the central point potential energy of each fuzzy model at the kth moment based on the error vector at the kth moment, the error vector at the k-1th moment, and the central point potential energy of each fuzzy model at the k-1th moment; Based on the central point potential energy of all the fuzzy models at the kth moment, determining the cluster center potential energy at the kth moment corresponding to the central error vector at the kth moment; Based on the cluster center potential energy at the k-th moment, reinitialize the center point potential energy of each fuzzy model at the k-th moment; Calculating the distance between the error vector at the kth moment and the new center point potential of each of the fuzzy models at the kth moment, and determining the shortest distance from the calculated multiple distances; A first update condition is constructed based on a preset potential energy upper bound, and a second update condition is constructed based on the preset potential energy upper bound, the preset potential energy lower bound, the cluster center potential energy at the first moment, and the shortest distance; if the potential energy of the error vector at the kth moment satisfies the first update condition, or the potential energy of the error vector at the kth moment and the cluster center potential energy at the kth moment satisfy the second update condition, then based on the central error vector at the kth moment, a first center point and a second center point are determined, and based on the first center point and the second center point, a Gaussian membership function of the new fuzzy model at the k+1th moment is constructed; Calculating the optimal control torque rate at the kth moment based on the error vector at the kth moment and the weight coefficient of each fuzzy model at the kth moment; Based on the error vector at the kth moment and the intermediate parameter matrix and consequent parameter sequence of each fuzzy model at the kth moment, respectively determining the intermediate parameter matrix of each fuzzy model at the k+1th moment; wherein the intermediate parameter matrix of each fuzzy model at the k+1th moment is used to update the consequent parameter sequence of each fuzzy model at the k+1th moment; Based on the error vector and optimal control torque rate at the kth moment and the intermediate parameter matrix and consequent parameter sequence of each fuzzy model at the kth moment, the consequent parameter sequence of each fuzzy model at the k+1th moment is determined respectively.

6. The method according to claim 5, characterized in that Based on the error vector at the kth moment, the formula for calculating the fusion term at the kth moment is as follows: Where, Representative The fusion term at each moment; Representative The error vector at each moment The vector values, ; The formula for calculating the fusion term at the first k-1 moments based on the error vector at the first k-1 moments is as follows: Where, Represents the fusion term of the first k-1 moments; For the The error vector at each moment The vector values, ; The formula for calculating the cross-fusion term at k moments based on the error vector at k moments is as follows: Where, represents the cross-fusion term of k moments; The formula for calculating the potential energy of the error vector at the kth moment based on the fusion term at the kth moment, the fusion term at the previous k-1 moments, and the cross-fusion term at the kth moment is as follows: Where, is the error vector at the kth moment potential energy; Among them, based on the error vector at the kth moment, the error vector at the k-1th moment, and the central point potential energy of each fuzzy model at the k-1th moment, the formula for calculating the central point potential energy of each fuzzy model at the kth moment is as follows: Where, For the The corresponding central error vector when the fuzzy model is constructed is: Representative The potential energy of the center point of a fuzzy model at the kth moment; Representative The potential energy of the center point of the fuzzy model at the k-1th moment; Representative The error vector at each moment The The vector value and The error vector at each moment The The distance between vector values; Among them, the cluster center potential energy at the kth moment is: , represents the cluster center potential energy at the kth moment; represents the number of fuzzy models, It is an initialization symbol, and its function is to assign values; is the central error vector at the kth moment, ; Wherein, based on the cluster center potential energy at the kth moment, the formula for resetting the center point potential energy of each fuzzy model at the kth moment is: Where, is a natural constant, , is a positive constant; The first update condition is: ; The second update condition is: ; is the preset upper bound of potential energy, is the preset potential energy lower bound, is the cluster center potential energy at the first moment, is the shortest distance, is a constant; The formulas for the first center point and the second center point are: In the formula, the new fuzzy model is A fuzzy model, is the first center point, is the second center point, is a fixed width; Among them, The formula of the Gaussian membership function of a fuzzy model at the k+1th moment is: Where, Representative The error vector at each moment The vector values; 、 Respectively The upper membership function and lower membership function of a fuzzy model; express Obey As the center point, is a Gaussian distribution with variance; express Obey As the center point, is a Gaussian distribution with variance; The formula for calculating the optimal control torque rate at the kth moment based on the error vector at the kth moment and the weight coefficient of each fuzzy model at the kth moment is as follows: Where, is the optimal control torque rate at the kth moment; For the The fuzzy model is The weight coefficient at each moment, For the The ideal optimal parameters of the fuzzy model are For the The inherent approximation error of the fuzzy model; Among them, calculate the The fuzzy model is The formula of the intermediate parameter matrix at each moment is as follows: Where, No. The fuzzy model is The intermediate parameter matrix at each moment; For the The fuzzy model is The intermediate parameter matrix at each moment, For the The fuzzy model is The consequent parameter sequence of the moment, ; Among them, calculate the The fuzzy model is The formula for the consequent parameter sequence at a moment is as follows: Where, For the The fuzzy model is The sequence of consequent parameters at each moment.

7. A complex dynamic trajectory tracking device based on incremental learning, characterized in that: A controller for a robotic arm, the device comprising: The acquisition module is used to obtain the joint drive data and torque control data of the previous moment; A compensation calculation module is used to determine the torque compensation data at the current moment based on the joint drive data at the previous moment and the track point on the target track that is closest to the tracking point at the current moment; The compensation calculation module is further configured to add the torque compensation data at the current moment to the torque control data at the previous moment to obtain the torque control data at the current moment; The drive calculation module is used to input the torque control data at the current moment into the pre-trained tracking control model to obtain the joint drive data at the current moment; a drive control module, configured to drive the respective rotational joints of the robotic arm to move using the joint drive data at a current moment, so that the end effector of the robotic arm tracks the target trajectory; The controller runs an interval type-2 evolving fuzzy neural network, which is updated at each moment when the end effector tracks the target trajectory; the compensation calculation module is specifically used to: Determining a tracking point of the end effector at a current moment based on joint drive data at a previous moment; Determining the track point closest to the current tracking point from the target track; determining a tracking error at a current moment based on the trajectory point and the tracking point; Obtain the tracking error at the previous moment, and subtract the current tracking error from the previous moment to obtain the error change rate at the current moment; Combining the error change rate and the tracking error at the current moment to obtain an error combination vector, and expanding the error combination vector to obtain an error vector at the current moment; The error vector at the current moment is input into the interval type-II evolutionary fuzzy neural network at the current moment to obtain the torque compensation data at the current moment.

8. A robotic arm, characterized in that: The robotic arm includes a controller, which is used to implement the complex dynamic trajectory tracking method based on incremental learning as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for complex dynamic trajectory tracking based on incremental learning according to any one of claims 1 to 6 is implemented.

Images