A robot dynamics parameter online identification method and system

Abstract

The application relates to the technical field of adaptive control of industrial robots, and provides a robot dynamics parameter online identification method and system, which comprises the following steps: obtaining a desired motion state and an actual motion state of an industrial robot, constructing tracking error information and dynamics regression information based on a robot dynamics model and linear parameterization characteristics; performing filtering processing on a dynamics regression matrix in the dynamics regression information, and constructing instantaneous information according to the filtered regression information; continuously performing pair-by-pair direction forgetting updating on historical regression information based on the instantaneous information, so as to obtain historical information reflecting a historical motion excitation characteristic of the robot; performing singular value decomposition processing on the historical information, and constructing parameter prediction error based on the spectrum-protected historical information; and jointly using the parameter prediction error and the tracking error information for online updating of robot dynamics parameters, and limiting the updating range of a dynamics parameter estimated value through bounded projection constraint.

Description

Technical Field

[0001] This invention relates to the field of adaptive control technology for industrial robots, specifically to a method and system for online identification of robot dynamic parameters. Background Technology

[0002] Industrial robots possess highly nonlinear and strongly coupled dynamic characteristics. During actual operation, they are also affected by factors such as load variations, joint friction, model parameter perturbations, and unmodeled dynamics, leading to uncertainties in robot dynamic parameters. Adaptive control can adjust the estimated dynamic parameters online based on the robot's real-time motion state, thereby reducing the impact of parameter uncertainty on control performance. However, traditional adaptive control parameter updates primarily rely on instantaneous tracking error information, making it highly dependent on the degree of excitation of the motion trajectory. When the robot is under conditions of weak excitation, intermittent excitation, or insufficient excitation in some parameter directions, the dynamic parameters are difficult to fully identify, thus affecting the model compensation accuracy and robot motion tracking performance.

[0003] To reduce the dependence of parameter learning on continuous excitation conditions, composite learning control constructs parameter prediction errors by introducing historical motion data of the robot. These prediction errors, along with real-time tracking error information, are used together to update dynamic parameters, thereby improving the utilization efficiency of historical information and parameter identification capabilities under limited excitation conditions. However, existing composite learning control still has certain shortcomings. On the one hand, the continuous accumulation of historical data may lead to the long-term retention of outdated information, while traditional overall forgetting methods may simultaneously weaken effective information that still has identification value. On the other hand, under weak excitation conditions, the historical information matrix may exhibit ill-conditioned or approximately rank-deficient behavior, and small singular values ​​are prone to generating large inverse gains during pseudo-inverse operations, thus amplifying the impact of measurement noise, regression errors, and unmodeled dynamics on the parameter learning process. Therefore, how to reasonably update historical information and suppress noise amplification in the weak spectral direction is a problem that urgently needs to be solved to improve the accuracy and reliability of parameter identification in industrial robot composite learning control. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for online identification of robot dynamic parameters. This method combines historical prediction errors and real-time tracking error information for online updating of dynamic parameters, and integrates a continuous pairwise forgetting mechanism and truncated singular value decomposition to achieve stable composite learning control of industrial robots under weak excitation conditions.

[0005] To achieve the above objectives, the following solution is adopted: In a first aspect, the present invention provides a method for online identification of robot dynamic parameters, comprising the following steps: Step S1: Obtain the desired motion state and actual motion state of the industrial robot, and construct tracking error information and dynamic regression information based on the linear parameterization model of robot dynamics; Step S2: Filter the dynamic regression matrix in the dynamic regression information, and construct instantaneous information based on the filtered regression information; Step S3: Based on the instantaneous information, perform continuous pairwise directional forgetting updates on the historical regression information to obtain historical information reflecting the historical motion excitation characteristics of the robot; Step S4: Perform truncated singular value decomposition on the historical information, and construct parameter prediction error based on the spectral-protected historical information; Step S5: Use the parameter prediction error and the tracking error information together for online updating of robot dynamic parameters, and limit the update range of dynamic parameter estimates by bounded projection constraints.

[0006] As a further technical solution, in step S1, the robot dynamics model is represented as follows:

[0007] in, , and These are the robot's joint position, joint velocity, and joint acceleration, respectively. The inertia matrix, The matrix represents the Coriolis force and the centrifugal force. For gravitational torque, For joint friction torque, For robot joint control input; Based on the robot's desired joint position Compared with actual joint position The tracking error information can be constructed, and the tracking error information includes position tracking error. e ( t and filtering tracking error s ( t ),as follows:

[0008]

[0009] in, To track the error gain matrix, auxiliary variables can be further constructed. .

[0010] Utilizing the linear parameterization property of robot dynamics, the linear parameterization model of robot dynamics can be expressed as:

[0011] in, Y r The dynamic regression matrix is ​​determined by the robot's motion state and the reference motion state. Θ The dynamics regression information consists of the robot dynamics parameter vector to be estimated, the dynamics regression matrix, and the dynamics parameter vector.

[0012] As a further technical solution, in step S2, the dynamic regression matrix is ​​filtered, specifically including: Decompose the dynamic regression matrix containing robot joint acceleration information:

[0013] in, and This is a known matrix determined based on the robot's dynamic parameterized structure.

[0014] The dynamic regression matrix and robot joint control inputs are subjected to stabilization filtering, and the direct dependence on robot joint acceleration information is eliminated by combining the decomposition relationship of the dynamic regression matrix, thus obtaining the filtered regression matrix. and filter joint control input And make it satisfy:

[0015] Based on the filtered regression matrix and the filtered joint control input, the instantaneous information used to characterize the excitation characteristics of the dynamic parameters in the current motion state of the robot is constructed as follows:

[0016]

[0017] The instantaneous information matrix h ( t This is used to characterize the excitation information of the current robot motion state in relation to the directions of each dynamic parameter, combined with the instantaneous information vector. b ( t This is used for updating and constructing subsequent historical information.

[0018] As a further technical solution, in step S3, the effective information direction corresponding to the robot's current motion state is determined based on the instantaneous information matrix, and the historical information is updated using a continuous pairwise direction forgetting method; the instantaneous information matrix is ​​subjected to feature decomposition, and a historical information matrix within the current information subspace is constructed:

[0019]

[0020] in, V ( t The column vectors constitute the instantaneous information matrix. h ( t The effective information subspace of ) The positive eigenvalues ​​corresponding to the subspace containing the effective information are: A ( t () is the representation matrix of existing historical information in the current valid information subspace; H ( t () represents the historical information matrix; Based on the effective information subspace, the historical information matrix H ( t ) and historical information vector B ( t Synchronize updates according to the same information direction:

[0021]

[0022] in, For directional forgetting rate, A + ( t )for A ( t The Moore-Penrose pseudo-inverse is obtained by synchronously updating and selectively decaying historical regression information that overlaps with the current effective information direction, while maintaining the consistency between the historical information matrix and the historical information vector in the direction forgetting process.

[0023] As a further technical solution, in step S4, the historical information matrix is ​​subjected to truncated singular value decomposition, which decomposes the historical information matrix into different spectral directions. Then, based on a preset spectral truncation threshold, selective inversion is performed on each spectral direction to construct a spectral-protected pseudo-inverse. as follows:

[0024]

[0025] in, h i The spectral values ​​of the historical information matrix, Q ( t ) represents the corresponding orthogonal eigenvector matrix. Spectral cutoff threshold; It is the spectral cutoff function; By retaining the effective information direction that meets the spectral truncation threshold and truncating the inverse operation of the weak spectral direction, the excessive inverse gain generated by the weak information direction during the pseudo-inverse of historical information is limited. Based on the aforementioned spectrum, protect against pseudo-inverse Historical Information Matrix H ( t ), historical information vector B ( t and current robot dynamics parameter estimates Constructing historical prediction errors :

[0026] As a further technical solution, the historical information matrix is ​​synchronously stopped when the robot's filtered regression information satisfies the finite interval excitation condition. H ( t ) and historical information vector B ( t The direction of the forgetting update is maintained, while retaining the historical information obtained when the finite interval excitation was established, thereby preserving the obtained effective parameter excitation information; the finite interval excitation condition satisfies:

[0027] in, Establish the timing for the finite interval incentive. For the length of the excitation interval, This serves as the lower bound for the incentive level. The spectral cutoff threshold Determined based on the excitation level and directional forgetting rate within a finite interval, and satisfying... This ensures that the effective parameter directions in the historical information matrix after the establishment of the finite interval excitation are all preserved by the truncated singular value decomposition, which is used for subsequent composite learning and updating of robot dynamic parameters.

[0028] As a further technical solution, in step S5, the tracking error information obtained based on the robot's real-time motion state and the aforementioned spectral protection historical prediction error are used together for the online update of the robot's dynamic parameters to construct a composite learning parameter update direction:

[0029] in, The online update rate of the dynamic parameter values, It is a positive definite adaptive gain matrix. For composite learning gain, The bounded projection operator updates the composite learning parameters, enabling the robot's real-time tracking information and historical effective stimulus information to jointly drive the dynamic parameter estimation, and restricts the estimated dynamic parameter values ​​to within the allowed set of dynamic parameters through the bounded projection operator.

[0030] As a further technical solution, the robot joint control input is constructed based on robot tracking error information, dynamic regression information, and the estimated dynamic parameters updated by bounded projection:

[0031] in, K D The positive definite feedback gain matrix is... The current estimated values ​​of the robot's dynamic parameters are used; the robot joint control input is applied to the robot joint drive mechanism to achieve closed-loop motion control of the robot by utilizing tracking error feedback and dynamic parameter estimation compensation.

[0032] Secondly, the present invention provides an online identification system for robot dynamic parameters, comprising: The first module is configured to acquire the desired motion state and actual motion state of the industrial robot, and construct tracking error information and dynamic regression information based on the robot dynamics model and linear parameterization features. The second module is configured to filter the dynamic regression matrix and construct instantaneous information based on the filtered regression information. The third module is configured to continuously update the historical regression information in pairs based on the instantaneous information to obtain historical information reflecting the historical motion excitation characteristics of the robot. The fourth module is configured to perform truncated singular value decomposition on the historical information and construct parameter prediction error based on the spectrally protected historical information. The fifth module is configured to use the parameter prediction error and the tracking error information together for online updating of robot dynamic parameters, and to limit the update range of the dynamic parameter estimates by bounded projection constraints.

[0033] Thirdly, the present invention provides an electronic device comprising: a processor and a memory; the memory is used to store program code and transmit the program code to the processor; the processor is used to execute the steps of the online identification method for robot dynamic parameters according to the instructions in the program code.

[0034] Fourthly, a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements a method for online identification of robot dynamic parameters.

[0035] Fifthly, the present invention provides a computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the steps of an online identification method for robot dynamic parameters.

[0036] Compared with the prior art, the advantages and positive effects of this invention are: 1. This invention incorporates real-time robot tracking information and historical prediction information into the dynamic parameter update process, making full use of the effective excitation information obtained during robot operation and reducing the dependence of parameter learning on continuous excitation conditions. After satisfying the excitation conditions in a finite interval, it can achieve exponential convergence of dynamic parameter errors and improve the convergence consistency of different parameter directions, thereby improving the dynamic compensation accuracy and motion tracking performance of industrial robots.

[0037] 2. This invention employs a continuous pairwise directional forgetting mechanism, which selectively updates historical information based on the effective information direction corresponding to the robot's current motion state. It only attenuates historical information that overlaps with the current information subspace and performs consistent directional transformation on the historical information matrix and historical information vector. This weakens the influence of outdated historical information while retaining information with parameter identification value in insufficiently excited directions, avoiding the synchronous attenuation of historical information in different parameter directions caused by traditional overall exponential forgetting, and improving the utilization efficiency and effectiveness of historical information.

[0038] 3. To address the problem that the historical information matrix is ​​prone to ill-conditioned or near-rank deficient conditions during the initial operation phase of the robot and under weak excitation conditions, this invention employs truncated singular value decomposition to protect the spectrum of the historical information matrix. Pseudo-inverse operations are performed only on effective spectral directions that reach a preset threshold, and excessive inverse gains generated by weak spectral directions are truncated. This suppresses the amplification of measurement noise, regression error, unmodeled dynamics, and numerical errors in historical prediction errors, thereby improving the numerical stability, noise resistance, and operational reliability of the industrial robot control system during the parameter learning process. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0040] Figure 1 This is a control block diagram of a composite learning industrial robot based on continuous pairwise directional forgetting and truncated singular value decomposition in an embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram of the structure of an N-degree-of-freedom industrial robot control system based on the composite learning control method described in this embodiment of the invention.

[0042] Figure 3 This is the trajectory tracking curve of the industrial robot in this embodiment of the invention.

[0043] Figure 4 This is the evolution trajectory of the dynamic parameters of the industrial robot in this embodiment of the invention. Detailed Implementation

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the following embodiments are used to explain the technical solutions of the present invention, and not to limit the scope of protection of the present invention. Without departing from the technical concept of the present invention, equivalent substitutions or adjustments made by those skilled in the art to the relevant control parameters and implementation forms based on the actual industrial robot structure, number of degrees of freedom, and control tasks should all fall within the scope of protection of the present invention.

[0045] Example 1 like Figure 1 and Figure 2 As shown, this embodiment uses a serial industrial robot with (N) controllable joints as the controlled object. The control system mainly includes a host computer control system, a joint drive system, a state feedback system, and an industrial robot with (N) degrees of freedom. The host computer control system interacts with the industrial robot through a communication interface to receive the robot's actual motion state and the given desired motion state, and runs a composite learning control program based on truncated singular value decomposition and continuous pairwise forgetting. The joint drive system drives the movement of each controllable joint of the industrial robot according to the joint control input output by the host computer. The state feedback system acquires the motion state of each joint of the robot in real time and feeds it back to the host computer, thereby forming a closed-loop control system.

[0046] During robot operation, the host computer first establishes tracking error information based on the robot's desired motion state and actual motion state, and combines this with the robot's dynamics model to obtain regression information for online parameter learning and dynamic compensation. Subsequently, the dynamic regression information is filtered to construct instantaneous parameter excitation information while avoiding direct use of joint acceleration measurements. Based on this instantaneous parameter excitation information, continuous pairwise directional forgetting is implemented on the obtained historical regression information, selectively attenuating outdated historical information overlapping with currently valid information directions, while retaining historical information with parameter identification value in currently insufficiently excited directions.

[0047] Based on this, the updated historical information matrix is ​​subjected to truncated singular value decomposition to retain effective information directions that meet the preset spectral threshold, while limiting excessive inverse gain generated by weak spectral directions during pseudo-inverse calculation. Furthermore, historical prediction errors are constructed using the spectrally protected historical information, and these errors, along with the robot's real-time tracking error information, are used for online updating of dynamic parameters. Simultaneously, bounded projection constraints are used to limit the range of parameter estimates to prevent situations where the parameters exceed the preset allowable range during estimation.

[0048] After updating the dynamic parameters, the host computer calculates the control inputs for each joint based on the robot's real-time tracking status, dynamic regression information, and the updated dynamic parameter estimates, and sends these joint control inputs to the joint drive system. Under the action of the joint drive system, the industrial robot executes the desired motion trajectory, and its actual motion state is fed back to the host computer. This process is repeated cyclically to achieve composite learning closed-loop control of the industrial robot under conditions of limited or insufficient excitation.

[0049] like Figure 2 As shown, the industrial robot control system in this embodiment includes a host computer control system, a joint drive system, a state feedback system, and a (N)-degree-of-freedom serial industrial robot. The industrial robot includes multiple robot links and (N) controllable joints connected in sequence. Each joint is driven by a corresponding joint drive unit to achieve independent or coordinated movement of each joint of the robot.

[0050] The host computer control system includes a processor, a memory, and a communication interface. The memory stores the control program for the composite learning robot described in this invention. The processor executes the control program and performs tracking error calculation, dynamic information processing, historical information updating, online estimation of dynamic parameters, and joint control input calculation based on the robot's desired and actual motion states. The host computer control system interacts with the joint drive system through the communication interface, sending the calculated robot joint control inputs to the corresponding joint drive units.

[0051] The state feedback system is used to acquire the actual motion state of the industrial robot. In this embodiment, the actual motion state of the robot includes at least the actual position and actual velocity of each joint, and the acquired actual motion state is fed back to the host computer control system in real time. For the subsequent filtering regression and composite learning process, the joint acceleration measurement value is not directly used as a necessary input. Instead, the robot joint position, joint velocity, and control input are used to construct the corresponding filtered dynamic regression information to reduce the noise influence introduced by direct measurement or numerical differentiation of joint acceleration.

[0052] Let the actual joint positions and joint velocities of the industrial robot be respectively... and ,in , Based on the robot's specific task, the host computer pre-defines or generates in real-time the desired joint motion trajectories, including the desired joint positions. Desired joint velocity and expected joint acceleration The desired motion state and the actual motion state are used together to construct the subsequent tracking error and reference motion state.

[0053] Within a control cycle, the state feedback system first collects the actual motion state of each joint of the robot and sends it to the host computer control system via the communication interface. The host computer reads the corresponding desired motion state, runs the composite learning control program, calculates the estimated values ​​of the robot's dynamic parameters and the control inputs of each joint, and then sends the joint control inputs to the joint drive system, which drives the movement of each robot joint. The updated actual motion state of the robot is again collected by the state detection device and fed back to the host computer, thus forming a real-time closed-loop control process for the industrial robot.

[0054] The specific hardware configuration of the host computer control system, joint drive system, and state feedback system can be configured according to the control interface of the industrial robot and the actual application requirements. As long as the robot's motion state can be collected, the control algorithm can be calculated in real time, and the joint control input can be transmitted, it can be used to implement the composite learning industrial robot control method described in this invention.

[0055] To construct a composite learning controller for industrial robots, this embodiment first establishes a dynamic model of an industrial robot with (N) degrees of freedom in series. The robot joint space dynamics are represented as follows:

[0056] in, , and These are the robot's joint position, joint velocity, and joint acceleration, respectively. The inertia matrix, The matrix represents the Coriolis force and the centrifugal force. For gravitational torque, For joint friction torque, This refers to the joint control input calculated by the host computer control system and sent to the robot joint drive system. The dynamic model corresponds to the subsequent parameter learning and dynamic compensation process of this invention.

[0057] Industrial robot dynamics exhibit linear parameterization characteristics, meaning that for a given auxiliary velocity vector... and its derivative There exists a known dynamic regression matrix determined by the robot's structural parameters and motion state, such that the unknown parameters in the robot's dynamics can be represented as a constant parameter vector, satisfying...

[0058] in, Let be the vector of robot dynamic parameters to be estimated online. p This refers to the number of parameters in the adopted dynamic parameterization model. The parameter vector may include combined parameters related to dynamic characteristics such as robot link mass, center of mass position, moment of inertia, and joint friction.

[0059] Define the robot position tracking error as And further construct the filtering tracking error ,in, This is a preset positive definite tracking error gain matrix, used to adjust the combined effect of position tracking error and velocity tracking error in the closed-loop control process.

[0060] Based on the position tracking error and the filtered tracking error, a reference velocity and a reference acceleration are constructed. , Therefore, we can obtain In other words, the filtered tracking error reflects the deviation between the robot's actual joint velocity and the reference velocity. Substituting the aforementioned reference velocity and reference acceleration into the robot's linear parameterization relationship, the dynamic regression matrix used for controller design is defined as follows: Therefore, the robot dynamics compensation term can be constructed using the known regression matrix Yr(t) and the estimated values ​​of the dynamic parameters.

[0061] In actual control processes, the robot's true dynamic parameters Since the parameters cannot be obtained directly, the estimated values ​​are maintained in real time by the host computer. And define the parameter estimation error as The parameter estimates will be updated online in subsequent implementation steps based on the robot's real-time tracking error information and historical prediction error information, and will be used for robot dynamics compensation and joint control input calculation.

[0062] Based on the linear parameterization characteristics of robot dynamics, when the auxiliary velocity is taken as the actual joint velocity of the robot, the robot dynamics can be expressed as:

[0063] in, This is the robot dynamics regression matrix. Since this regression matrix includes joint accelerations, directly using it for online parameter learning would require additional acquisition of joint acceleration signals. Therefore, based on the structure of the basis functions of each parameter in the robot dynamics regression matrix, [the following is omitted as it is not directly related to the previous sentence].Y r Decomposed into

[0064] in, and The regression matrix, pre-determined based on the robot's known dynamic structure, includes all terms related to joint acceleration. Y a In the time derivative.

[0065] Set filter constants A first-order linear filter is used to filter both the robot joint control input and the regression matrix. The filtered joint control input is defined as follows: and the filter state matrix ,satisfy

[0066] The initial value of the corresponding filter state is set to zero. Filter constant. The dynamic response speed of the filter is adjusted, and its value can be selected based on the robot's operating frequency, the noise level of the measured signal, and the bandwidth of the control system. Y a By filtering the time derivative term and applying integration by parts, a filtered regression matrix that does not explicitly contain robot joint accelerations can be obtained.

[0067] By combining the filtered joint control input, the filtered dynamic regression relationship can be further obtained.

[0068] Therefore, the information related to joint acceleration in the original dynamic model is converted into filtered regression information composed of robot joint position, joint velocity, joint control input, and filter internal state. Within each control cycle, the host computer calculates based on the currently acquired robot joint position and joint velocity. Y a and Y b Synchronously update filter status and filter joint control input Thus, the filtered regression matrix is ​​obtained. Based on the above information, an instantaneous information matrix can be defined. and instantaneous information vector for

[0069] Instantaneous information matrix h ( tThe instantaneous information vector is used to characterize the excitation information obtained by the robot in different dynamic parameter directions under the current motion state. b ( t This is used to establish the correspondence between the current motion data and the dynamic parameters to be identified. Then there exists...

[0070] At the same time, it can be seen from the construction method of the instantaneous information matrix that That is, the instantaneous information matrix is ​​a symmetric positive semi-definite matrix. Therefore, h ( t The non-zero eigenvalues ​​and their corresponding eigenvectors of the robot can reflect the effective excitation direction of the parameters actually obtained under the current motion state, while the parameter direction corresponding to zero eigenvalues ​​or weaker eigenvalues ​​indicates that the current motion state provides weak or no effective excitation to the corresponding dynamic parameters.

[0071] During the actual operation of the robot, the host computer control system performs the following adjustments in each control cycle based on the current filter regression matrix: Y f ( t ) and filter joint control input Real-time updates h ( t )and b ( t Therefore, instantaneous information can dynamically reflect the parametric excitation characteristics provided by the current trajectory as the robot's motion state changes, without the need to store all the original robot motion data for a long time.

[0072] When the rank of h(t) satisfies r(t)>0, perform minimum singular value decomposition on h(t):

[0073] in, Each column is an orthogonal eigenvector corresponding to a non-zero eigenvalue, which is used to construct an effective information subspace of the robot's current motion state; Let be a diagonal matrix composed of corresponding positive eigenvalues. Further define the historical information matrix within the current effective information subspace as .

[0074] in, This is a historical information matrix. Therefore, Used to determine the direction of the dynamic parameters actually excited by the current robot motion trajectory. This is used to characterize the distribution of existing historical information in the current effective information subspace.

[0075] Before establishing the excitation conditions in the finite interval, a directional forgetting rate is set. Historical information matrix H ( t ) and historical information vector B ( t Update synchronously according to the following formula:

[0076] Where A+ is the Moore-Penrose pseudoinverse of A. This is the historical information vector corresponding to the historical information matrix H(t). When r(t) = 0, it indicates that the current motion state does not provide new effective parameter excitation information, and the forgetting terms in both directions mentioned above are zero.

[0077] In the above update methods, instantaneous information h ( t )and b ( t This is used to continuously write new parameter information obtained in the current robot motion state into the historical information; at the same time, the direction forgetting term only acts on the historical information directions that overlap with the current effective information subspace. Thus, when a certain dynamic parameter direction receives a new effective excitation at the current moment, the earlier historical information obtained in that direction can be gradually attenuated and updated using the current information; for parameter directions that are not currently sufficiently excited, their existing historical information is not attenuated as a whole, so that previously obtained information that still has parameter identification value can continue to be retained.

[0078] Compared to the overall forgetting method that applies exponential decay to all historical information, this embodiment adopts selective directional forgetting based on the current information subspace, so that the update of historical information corresponds to the current parameter excitation direction of the robot, thereby reducing the loss of information in irrelevant directions and improving the utilization efficiency of historical data under limited excitation and intermittent excitation conditions.

[0079] Furthermore, this embodiment includes a historical information matrix. H ( t ) and historical information vector B ( t Using the same information subspace and direction transformation for pairwise updates, then under ideal noise-free conditions, the constructed historical information pairs can maintain...

[0080] This provides a consistent regression basis for subsequently using historical information to construct dynamic parameter prediction errors. In the actual control process, the host computer first uses the instantaneous information matrix within each control cycle... h ( t Determine the current valid information subspace and update the historical information matrix synchronously. H ( t) and historical information vector B ( t As the robot's trajectory and excitation direction change, the historical information can be continuously and selectively updated in a directional manner, so that earlier historical information that overlaps with the current effective information direction gradually decays, while retaining existing effective information in the current insufficiently excitation direction.

[0081] Due to historical information matrix H ( t If is a symmetric positive semi-definite matrix, its singular value decomposition can be equivalently represented as its eigenvalue decomposition:

[0082] in, It is an orthogonal matrix, and each column is an orthogonal eigenvector corresponding to the historical information matrix; h i ( t Let be the eigenvalues ​​of the historical information matrix in different parameter directions, which are also its singular values, and satisfy the following conditions: .

[0083] Eigenvalues ​​are used to characterize the degree of effective excitation of historical motion data in the direction of the corresponding dynamic parameter. When a certain h i ( t When the value is small, it indicates that the robot's existing historical motion data has weak excitation information in the corresponding parameter direction; if a normal Moore-Penrose pseudo-inverse is used directly, the inverse gain in the corresponding direction is 1 / h i ( t ), easy to h i ( t When the gain is small, it produces a large inverse gain, thereby amplifying measurement noise, regression error, and numerical error.

[0084] Therefore, this embodiment sets a spectral cutoff threshold. And define the spectral cutoff function.

[0085] Therefore, a pseudo-inverse of the historical information matrix after truncated singular value decomposition is constructed:

[0086] in, h i The spectral values ​​of the historical information matrix, Q ( t ) represents the corresponding orthogonal eigenvector matrix. Spectral cutoff threshold; It is the spectral cutoff function; When the spectral value corresponding to a certain historical information direction is greater than or equal to the preset spectral truncation threshold When the direction is retained, the inverse is performed according to its spectral value; when the corresponding spectral value is less than When this happens, the inverse gain in that direction is set to zero, so that the direction of weak information is temporarily excluded from the calculation of historical prediction error. Thus, the maximum gain of the spectral-protected pseudo-inverse is limited by the spectral truncation threshold, which avoids excessive inverse gain from ordinary pseudo-inverses when the historical information matrix is ​​approximately rank-deficient.

[0087] Furthermore, in order to evaluate the current dynamic parameter estimation results using historical information, the historical information matrix is ​​first used as a basis. H ( t ), historical information vector B ( t and current dynamic parameter estimates. Constructing historical regressive residuals:

[0088] This characterizes the degree of inconsistency between the current estimated dynamic parameters and the robot's existing historical regression information.

[0089] By combining the historical regression residuals with the pseudo-inverse obtained from truncated singular value decomposition, a spectrum-protected historical prediction error is constructed:

[0090] The prediction error is not directly applied using unprocessed historical regression residuals. Instead, spectral selection is performed based on the effective excitation degree of each parameter direction in the historical information matrix, retaining only parameter directions with sufficient historical excitation information for parameter learning. Under ideal noise-free conditions, there exists...

[0091] in, This represents the estimation error of the dynamic parameters. Therefore, the spectral protection history prediction error essentially characterizes the component of the dynamic parameter estimation error in the direction of information preservation in the truncated singular value decomposition, enabling the parameter directions that have obtained sufficient historical excitation to continue to participate in composite learning, while the weak spectral directions with insufficient excitation in the current historical data do not temporarily generate high-gain parameter corrections.

[0092] To ensure that historical information acquired through continuous pairwise forgetting is stably retained after the robot has accumulated sufficient parameter excitation information, and to prevent changes in the already acquired valid information due to continuous updates, this embodiment introduces a finite interval excitation judgment mechanism during the historical information update process. When the robot's filtered regression information meets the finite interval excitation condition within a preset time interval, the historical information matrix is ​​stopped. H ( t ) and historical information vector B (t The information is continuously updated and kept as historical information when the finite interval incentives were established.

[0093] Specifically, if there is a finite time length t d >0. Finite interval excitation establishment time and the lower bound of the incentive level , so that the filtering regression matrix Y f ( t )satisfy

[0094] It is then assumed that the robot's filtered regression information is within the interval The condition satisfies the excitation condition over a finite interval, where, I p for p An identity matrix of order 1. This condition indicates that the motion data generated by the robot within the finite time interval can provide effective excitation information for the orientation of all dynamic parameters to be estimated.

[0095] exist At that time, the historical information matrix and historical information vector are continuously updated according to the continuous pairwise forgetting law, so as to continuously accumulate new instantaneous information generated by the robot's current motion state and selectively adjust existing historical information. When the excitation conditions in the finite interval are... Once the time frame is established, the historical information matrix and historical information vector are simultaneously frozen, i.e. .

[0096] Therefore, for The historical information matrix and historical information vector are respectively maintained as

[0097] The synchronous freezing allows the effective historical information already obtained by the robot in the limited excitation phase to be continuously used for parameter learning in the subsequent control process, while avoiding further changes to the obtained historical information in the subsequent weak excitation or no excitation operation phase.

[0098] Based on the update law of continuous pairwise forgetting historical information, when the excitation condition of the finite interval is met, the historical information matrix at the frozen time is... Satisfy the lower bound of positive definiteness Thus, even if the orientation forgetting mechanism is continuously employed before the establishment of the finite-interval excitation, the parameter information acquired by the robot within the finite excitation time interval can still form a positive definite historical information matrix. Simultaneously, to ensure that the truncated singular value decomposition mechanism retains all effective dynamic parameter orientations after the establishment of the finite-interval excitation, this embodiment further requires a spectral truncation threshold. Matching the incentive level within a finite interval, i.e. .

[0099] Due to the freezing of the historical information matrix All eigenvalues ​​are not less than When the spectral truncation threshold meets the above conditions, all parameter directions of the historical information matrix after the finite interval excitation is established are higher than the spectral truncation threshold. Therefore, truncated singular value decomposition will no longer truncate the parameter directions that have already received sufficient excitation. At this time, the historical information can continue to participate fully in the subsequent composite learning parameter updates.

[0100] The threshold compatibility conditions mentioned above are mainly used to ensure that all dynamic parameter directions participate in parameter learning after the excitation in the finite interval is established. If the selected spectral truncation threshold does not meet the above conditions, the truncated singular value decomposition can still limit the inverse gain of the weak spectral direction, but the parameter directions that are truncated by the spectrum will not participate in the complete historical prediction error calculation. Therefore, parameter learning can only be performed using the information directions retained by the truncated singular value decomposition.

[0101] During the actual control process, the host computer synchronously monitors the parameter excitation level provided by the filtered regression information during the continuous updating phase of historical information; when it detects that the filtered regression information within a preset finite time interval meets the above-mentioned finite interval excitation conditions, it updates the current historical information matrix. and historical information vector Save the data and stop subsequent updates for both. Then, the prediction error after truncated singular value decomposition continuously participates in the online learning of robot dynamic parameters using the frozen historical information, thereby enabling the reuse of finite excitation information in subsequent control processes.

[0102] Furthermore, the robot's real-time tracking error information is fused with historical prediction error information to drive online updates of dynamic parameters, thereby simultaneously utilizing the robot's current motion state and historical effective excitation information to improve parameter learning capabilities.

[0103] Specifically, a positive definite adaptive gain matrix is ​​set. And set the composite learning gain. The update law for the estimated dynamic parameters is as follows:

[0104] in, The online update rate of the dynamic parameter values; This is a real-time adaptive update term formed based on the robot's current motion state and tracking error, used to adjust the estimated values ​​of dynamic parameters according to the current control error; This is a composite learning update term formed based on historical effective stimulus information, used to continuously correct the estimated values ​​of dynamic parameters using historical information accumulated during the robot's previous operation. By leveraging the combined effects of these two types of information, parameter updates can avoid relying solely on the tracking error information at the current moment, and the efficiency of utilizing historical motion information under limited stimulus conditions can be improved. This refers to a projection operator constructed for the set of allowed parameters, preventing measurement noise, unmodeled dynamics, or transient errors from causing the estimated dynamic parameters to exceed the physically reasonable range.

[0105] After completing the dynamic parameter update, the current parameter estimates will be... Used for robot dynamics compensation. Combined with position tracking error. e ( t ), filtering tracking error s ( t and dynamic regression matrix Y r ( t Constructing robot joint control inputs

[0106] in, This is the positive definite feedback gain matrix. The position tracking error term and the filtered tracking error term are used to suppress the deviation between the robot's actual motion state and the desired motion state; the dynamic compensation term... The dynamic parameter estimation results obtained from composite learning are then used to perform online compensation for dynamic effects such as robot inertia, Coriolis force, centrifugal force, gravity, and joint friction. As the parameter estimates are continuously updated, historical prediction information and real-time tracking information jointly participate in the identification of dynamic parameters, thereby gradually improving the robot's dynamic compensation effect.

[0107] Following the proposed composite learning industrial robot control method based on spectral protection and continuous pairwise directional forgetting, the system executes the control cyclically according to a preset control cycle in the host computer control system. The overall control process is as follows: 1. Obtain the desired joint position, desired joint velocity, and desired joint acceleration of the industrial robot, and have the actual joint position and joint velocity of the robot collected by the state detection device. Calculate the position tracking error, filtered tracking error, and dynamic regression matrix for controller design based on the desired motion state and the actual motion state. 2. Based on the linear parameterization relationship of robot dynamics, perform stabilization filtering on the dynamic regression matrix and robot joint control inputs, update the corresponding filtered states, and obtain the filtered regression matrix. Y f ( t ) and filter joint control input This avoids directly using robot joint acceleration measurements during parameter learning; 3. Calculate the instantaneous information matrix and instantaneous information vector based on the filter regression matrix and the filter joint control input. h ( t and instantaneous information vector b ( t This allows us to obtain the excitation information of the dynamic parameters contained in the robot's current motion state; 4. Based on the instantaneous information matrix h ( t The effective feature directions are used to determine the current information subspace, and the instantaneous information is used to modify the historical information matrix. H ( t ) and historical information vector B ( t ) Perform continuous pairwise directional forgetting updates so that new effective parameter information is continuously written into historical information, while selectively decaying earlier historical information that overlaps with the current effective information direction; 5. Update the historical information matrix H ( t Perform spectral decomposition and cut off the spectrum according to the preset spectral truncation threshold. Construct a pseudo-inverse after truncating singular value decomposition; perform pseudo-inverse operation on effective information directions that reach the spectral truncation threshold, and set the inverse gain of weak information directions below the spectral truncation threshold to zero, so as to suppress the amplification of noise and regression error by excessive pseudo-inverse gain under weak excitation conditions. 6. Calculate the historical prediction error based on the spectral protection pseudo-inverse, historical information matrix, historical information vector, and current dynamic parameter estimates, and use the historical prediction error as historical learning information to participate in the online updating of dynamic parameters; 7. During the continuous updating of historical information, the excitation level of the filtered regression information is judged; when the filtered regression information within a preset finite time interval is detected to meet the finite interval excitation condition, the current historical information matrix and historical information vector are frozen synchronously, and the frozen effective historical information is used for parameter learning in the subsequent control process; the spectral truncation threshold should be lower than the effective spectrum lower bound of the frozen historical information matrix to avoid the truncated singular value decomposition of the dynamic parameter direction that has been sufficiently excited; 8. The robot real-time tracking error driving term and the spectrum protection historical prediction error driving term are used together to construct the composite learning parameter update direction, and the dynamic parameter estimate is updated online through the bounded projection operator to keep the parameter estimate within the preset allowable range; 9. Substitute the updated dynamic parameter estimates into the robot control law, calculate the control inputs for each joint of the robot based on the tracking error information and dynamic compensation information, and send the joint control inputs to the joint drive system through the host computer communication interface; 10. The industrial robot moves under the action of the joint drive system. The state feedback system re-collects the actual motion state of the robot and feeds it back to the host computer. Then, it enters the next control cycle and repeats the above steps.

[0108] Example 2 To further verify the effectiveness of the continuous pairwise orientation forgetting mechanism and the truncated singular value decomposition mechanism described in this invention, this embodiment establishes a six-degree-of-freedom serial industrial robot dynamics simulation model in the MATLAB / Simulink environment, and compares and verifies the continuous pairwise orientation forgetting mechanism and the TSVD spectrum protection mechanism respectively. Each set of comparative simulations uses the same robot dynamics model, controller parameters, initial estimates of dynamic parameters, and sampling period, only changing the corresponding control mechanism to be verified, in order to reduce the influence of other factors on the comparison results.

[0109] Establish robot base coordinate system In this coordinate system, the origin O0 is fixed to the robot base, the z0 axis is vertically upward along the robot base, the x0 axis is set directly in front of the robot's initial workspace, and the y0 axis is determined according to the right-hand coordinate system. Therefore, the motion of the end effector along the y-axis described in this embodiment refers to the robot end effector relative to a fixed base coordinate system. Translational motion along the y0 axis, rather than the local y-axis direction of the tool coordinate system that changes with the robot's end effector posture.

[0110] Let the position vector of the robot's end effector relative to the base coordinate system be... The distance the end effector moves along the y0 axis is defined as... To verify the ability of the continuous pairwise forgetting mechanism to utilize intermittent excitation information, three finite motions were set within a 20s simulation time, with sufficient intervals between each motion to simulate weak industrial excitation conditions.

[0111] like Figure 3 As shown, during the three intermittent excitation processes, the controller employing the continuous pairwise directional forgetting mechanism maintained good end-effector y0 displacement tracking performance, with the actual trajectory closely matching the desired trajectory. Without the continuous pairwise directional forgetting mechanism, the tracking performance in the first excitation stage was essentially the same as the method of this invention, but the tracking error gradually increased in the second and third excitation stages. This indicates that the continuous pairwise directional forgetting mechanism can improve the effective utilization of historical information, enabling the controller to maintain relatively stable trajectory tracking performance under intermittent excitation and changing operating conditions.

[0112] like Figure 4 As shown, during the initial operation and weak excitation phases of the robot, without truncated singular value decomposition (SVD), some dynamic parameter estimation trajectories exhibit significant transient fluctuations. With truncated SVD, the parameter estimation process becomes smoother, and initial peaks and parameter fluctuations are effectively suppressed. As the effective excitation information gradually increases, the parameter estimation results of both methods gradually stabilize. This indicates that the truncated SVD mechanism can suppress excessive inverse gain generated in the weak spectral direction, improving the numerical stability and noise robustness of the composite learning parameter update process.

[0113] This invention does not limit the specific number of degrees of freedom, control cycle, host computer model, or communication method of the industrial robot. As long as the robot motion state acquisition, historical information update, spectrum protection prediction error calculation, online learning of dynamic parameters, and joint control input generation can be achieved according to the above method, the composite learning industrial robot control method described in this invention can be implemented.

[0114] Example 3 This embodiment provides an online identification system for robot dynamic parameters, including: The first module is configured to acquire the desired motion state and actual motion state of the industrial robot, and construct tracking error information and dynamic regression information based on the robot dynamics model and linear parameterization features. The second module is configured to filter the dynamic regression matrix and construct instantaneous information based on the filtered regression information. The third module is configured to continuously update the historical regression information in pairs based on the instantaneous information to obtain historical information reflecting the historical motion excitation characteristics of the robot. The fourth module is configured to perform truncated singular value decomposition on the historical information and construct parameter prediction error based on the spectrally protected historical information. The fifth module is configured to use the parameter prediction error and the tracking error information together for online updating of robot dynamic parameters, and to limit the update range of the dynamic parameter estimates by bounded projection constraints.

[0115] Example 4 This embodiment provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs the steps of the online identification method for robot dynamic parameters in Embodiment 1.

[0116] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0117] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0118] Example 5 This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed, implements the online identification method for robot dynamic parameters in Embodiment 1.

[0119] Example 6 This embodiment provides an electronic device, which is intended to represent various forms of digital computers or various forms of mobile devices; the electronic device includes: a processor and a memory; the memory is used to store program code and transmit the program code to the processor; the processor is used to execute the steps of the online identification method for robot dynamic parameters in Embodiment 1 according to the instructions in the program code.

[0120] The components, their connections and relationships, and their functions shown in this embodiment are merely examples and are not intended to limit the implementation of the invention described and / or claimed in this embodiment.

[0121] An electronic device includes at least one processor and memory, such as read-only memory (ROM) or random access memory (RAM), communicatively connected to the processor. The memory stores computer programs executable by the processor. The processor can perform various appropriate actions and processes based on the computer programs stored in the ROM or loaded from memory locations into the RAM. The RAM may also store various programs and data required for the operation of the electronic device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0122] Multiple components in an electronic device are connected to an I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0123] A processor can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processors include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor performs the various methods and processes described above, such as image depth estimation methods.

[0124] In some embodiments, the image depth estimation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded into and / or installed on an electronic device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by a processor, one or more steps of the image depth estimation method described above may be performed. Alternatively, in other embodiments, the processor may be configured to perform the image depth estimation method by any other suitable means (e.g., by means of firmware).

[0125] Various implementations of the systems and technologies described above in this embodiment can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0126] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0127] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0128] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0129] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0130] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0131] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved. This embodiment does not impose any limitations on these steps.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for online identification of robot dynamic parameters, characterized in that, Includes the following steps: Step S1: Obtain the desired motion state and actual motion state of the industrial robot, and construct tracking error information and dynamic regression information based on the robot dynamics model and linear parameterization features. Step S2: Filter the dynamic regression matrix in the dynamic regression information, and construct instantaneous information based on the filtered regression information; Step S3: Based on the instantaneous information, perform continuous pairwise directional forgetting updates on the historical regression information to obtain historical information reflecting the historical motion excitation characteristics of the robot; Step S4: Perform truncated singular value decomposition on the historical information, and construct parameter prediction error based on the spectral-protected historical information; Step S5: Use the parameter prediction error and the tracking error information together for online updating of robot dynamic parameters, and limit the update range of dynamic parameter estimates by bounded projection constraints.

2. The online identification method for robot dynamic parameters as described in claim 1, characterized in that, In step S1, the robot dynamics model is represented as follows: in, , and These are the robot's joint position, joint velocity, and joint acceleration, respectively. The inertia matrix, The matrix represents the Coriolis force and the centrifugal force. For gravitational torque, For joint friction torque, For robot joint control input; Based on the robot's desired joint position Compared with actual joint position The tracking error information is constructed, and the tracking error information includes position tracking error. e ( t and filtering tracking error s ( t The specific formula is: in, The tracking error gain matrix; Constructing auxiliary variables ; Utilizing the linear parameterization property of robot dynamics, the robot dynamics model can be expressed as: in, Y r The dynamic regression matrix is ​​determined by the robot's motion state and the reference motion state. Θ The dynamics regression information consists of the robot dynamics parameter vector to be estimated, the dynamics regression matrix, and the dynamics parameter vector.

3. The online identification method for robot dynamic parameters as described in claim 2, characterized in that, The specific process of step S2 is as follows: Decompose the dynamic regression matrix containing robot joint acceleration information: in, and The known matrix is ​​determined based on the parameterized structure of the robot's dynamics; The dynamic regression matrix and robot joint control inputs are subjected to stabilization filtering, and the direct dependence on robot joint acceleration information is eliminated by combining the decomposition relationship of the dynamic regression matrix, thus obtaining the filtered regression matrix. and filter joint control input And make it satisfy: Based on the filtered regression matrix and the filtered joint control input, the instantaneous information used to characterize the excitation characteristics of the dynamic parameters in the current motion state of the robot is constructed as follows: in, h ( t ) represents the instantaneous information matrix; b ( t ) represents the instantaneous information vector; Instantaneous information matrix h ( t This is used to characterize the excitation information of the current robot motion state in relation to the directions of each dynamic parameter, combined with the instantaneous information vector. b ( t This is used for updating and constructing subsequent historical information.

4. The online identification method for robot dynamic parameters as described in claim 3, characterized in that, The specific process of step S3 is as follows: Based on the instantaneous information matrix, the effective information direction corresponding to the robot's current motion state is determined, and the historical information is updated using a continuous pairwise direction forgetting method; the instantaneous information matrix is ​​subjected to feature decomposition, and a historical information matrix within the current information subspace is constructed: in, V ( t The column vectors constitute the instantaneous information matrix. h ( t The effective information subspace of ) The positive eigenvalues ​​corresponding to the subspace containing the effective information are: A ( t () is the representation matrix of existing historical information in the current valid information subspace; H ( t () represents the historical information matrix; Based on the effective information subspace, the historical information matrix H ( t ) and historical information vector B ( t Synchronize updates according to the same information direction: in, For directional forgetting rate, A + ( t )for A ( t The Moore-Penrose pseudo-inverse is obtained by synchronously updating and selectively decaying historical regression information that overlaps with the current effective information direction, while maintaining the consistency between the historical information matrix and the historical information vector in the direction forgetting process.

5. The online identification method for robot dynamic parameters as described in claim 4, characterized in that, The specific process of step S4 is as follows: The historical information matrix is ​​subjected to truncated singular value decomposition, which decomposes the historical information matrix into different spectral directions. Then, based on a preset spectral truncation threshold, selective inversion is performed on each spectral direction to construct a spectral-protected pseudo-inverse. as follows: in, h i The spectral values ​​of the historical information matrix, Q ( t ) represents the corresponding orthogonal eigenvector matrix. Spectral cutoff threshold; It is the spectral cutoff function; By retaining the effective information directions that meet the spectral truncation threshold and truncating the inverse operation of the weak spectral directions, the excessive inverse gain generated by the weak information directions during the pseudo-inverse of historical information is limited. Based on the aforementioned spectrum, protect against pseudo-inverse Historical Information Matrix H ( t ), historical information vector B ( t and current robot dynamics parameter estimates Constructing historical prediction errors : When the robot's filtered regression information satisfies the excitation conditions within a finite interval, the historical information matrix is ​​synchronously stopped. H ( t ) and historical information vector B ( t The direction of the forgetting update is maintained, while retaining the historical information obtained when the finite interval excitation was established, thereby preserving the obtained effective parameter excitation information; the finite interval excitation condition satisfies: in, Establish the time for the finite interval excitation. For the length of the excitation interval, This serves as the lower bound for the incentive level. The spectral cutoff threshold Determined based on the excitation level and directional forgetting rate within a finite interval, and satisfying... This ensures that the directions of all effective parameters in the historical information matrix after the establishment of the finite interval excitation are preserved by the truncated singular value decomposition, which is then used for subsequent composite learning and updating of robot dynamic parameters.

6. The online identification method for robot dynamic parameters as described in claim 1, characterized in that, In step 5, The tracking error information obtained based on the robot's real-time motion state, together with the spectral protection historical prediction error, is used for the online updating of the robot's dynamic parameters. The update rules for the estimated dynamic parameters are as follows: in, The online update rate of the dynamic parameter values. It is a positive definite adaptive gain matrix. Here, p is the composite learning gain, and p is the bounded projection operator. Through the composite learning parameter update, the robot's real-time tracking information and historical effective stimulus information jointly drive the dynamic parameter estimation, and the bounded projection operator restricts the dynamic parameter estimates to within the allowed set of dynamic parameters.

7. An online identification system for robot dynamic parameters, characterized in that, include: The first module is configured to acquire the desired motion state and actual motion state of the industrial robot, and construct tracking error information and dynamic regression information based on the robot dynamics model and linear parameterization features. The second module is configured to filter the dynamic regression matrix and construct instantaneous information based on the filtered regression information. The third module is configured to continuously update the historical regression information in pairs based on the instantaneous information to obtain historical information reflecting the historical motion excitation characteristics of the robot. The fourth module is configured to perform truncated singular value decomposition on the historical information and construct parameter prediction error based on the spectrally protected historical information. The fifth module is configured to use the parameter prediction error and the tracking error information together for online updating of robot dynamic parameters, and to limit the update range of the dynamic parameter estimates by bounded projection constraints.

8. An electronic device, the electronic device comprising: A processor and a memory; the memory is used to store program code and transmit the program code to the processor; characterized in that the processor is used to execute the steps of the online identification method for robot dynamic parameters according to any one of claims 1-6 according to the instructions in the program code.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the online identification method for robot dynamic parameters as described in any one of claims 1-6.

10. A computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, When the program instructions are executed by a computer, the computer performs the steps of the online identification method for robot dynamic parameters according to any one of claims 1-6.

Images