Robot driving control method, device and equipment and storage medium

Abstract

The invention provides a robot driving control method and device, equipment and a storage medium. The robot driving control method comprises the steps that tracking errors are calculated according to expected trajectories and actual moving trajectories corresponding to all joints of a connecting rod robot at the current moment; calculating the tracking error by adopting a sliding mode control algorithm to obtain a first input torque at the current moment; taking the actual input torque at the previous moment and the actual moving track as the input of a dynamic neural network, predicting the unknown torque in the robot dynamic model at the current moment, and taking the predicted unknown torque as the second input torque at the current moment; the actual input torque at the current moment is determined according to the first input torque and the second input torque, and the connecting rod robot is controlled to move according to the actual input torque at the current moment; the problem that in the prior art, stability and accuracy of a connecting rod robot are reduced due to various uncertain factors is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot control, and particularly relates to a robot driving control method and device, equipment and a storage medium. BACKGROUND

[0002] Robots have been widely used in industrial assembly, robot welding, safety and explosion prevention, engineering machinery and other fields due to their unique operation and flexibility. For different task modes, different spatial motion trajectory planning needs to be performed on the connecting rod robot, so as to achieve accurate positioning or tracking. Therefore, it is particularly important to accurately, quickly and efficiently control the actuator of the connecting rod robot. The spatial trajectory tracking control problem of the connecting rod robot is to track the given expected trajectory by giving the driving torque between each connecting rod, so as to track the position, speed, angular displacement and other state information of the connecting rod. However, due to the structural uncertainty, non-structural uncertainty and environmental uncertainty in the modeling of the robot, the stability and accuracy of the robot will decrease, and eventually the trajectory of the robot during movement will deviate from the expected trajectory. SUMMARY

[0003] The present application provides a robot driving control method, device, equipment and storage medium, which is used to solve the problem of stability and accuracy decrease of the connecting rod robot caused by various uncertain factors in the prior art.

[0004] In a first aspect, the present application provides a robot driving control method, comprising: calculating a tracking error according to an expected trajectory and an actual running trajectory corresponding to each joint of the connecting rod robot at a current time; calculating the tracking error by using a sliding mode control algorithm to obtain a first input torque at the current time; taking the actual input torque at the previous time and the actual running trajectory as inputs of a dynamic neural network, predicting an unknown torque in a robot dynamics model at the current time, and taking the predicted unknown torque as a second input torque at the current time; determining an actual input torque at the current time according to the first input torque and the second input torque, and controlling the connecting rod robot to move according to the actual input torque at the current time.

[0005] In one or more possible embodiments, the taking the actual input torque at the previous time and the actual running trajectory as inputs of a dynamic neural network, predicting an unknown torque in a robot dynamics model at the current time, and taking the predicted unknown torque as a second input torque at the current time comprises: inputting an input vector Z as an input of a dynamic neural network; wherein the input vector Z comprises a system state, the tracking error and an actual input torque at a previous time; the system state is used to represent an actual running trajectory at a current time; performing nonlinear transformation on the input vector Z through a weight and a hidden layer activation function of the dynamic neural network to obtain an approximation output of the unknown torque, and taking the approximation output as a second input torque at the current time; wherein the approximation output contains an approximation error, and the approximation error is not greater than a preset value.

[0006] In one or more possible embodiments, further comprising: updating the weight of the dynamic neural network according to a preset update rate, the update rate being related to the tracking error and the actual input torque at the previous time.

[0007] In one or more possible embodiments, the preset update rate is calculated according to the following formula:

[0008] wherein, represents the preset update rate, is a diagonal matrix gain, j represents a number of hidden layer nodes, represents a hidden layer activation function, s is a sliding mode surface vector constructed according to the tracking error, represents a weight of the dynamic neural network, represents a normal number damping coefficient.

[0009] In one or more possible embodiments, the tracking error is calculated by using a sliding mode control algorithm to obtain the first input torque at the current time, comprising: determining a filtered error according to the tracking error and a derivative of the tracking error; constructing a non-singular terminal sliding mode surface based on the filtered error signal; wherein the non-singular terminal sliding mode surface is used to represent a deviation degree between a desired trajectory and an actual running trajectory of the articulated robot; calculating the first input torque at the current time according to the filtered error signal and the non-singular terminal sliding mode surface by using the following formula:

[0010] wherein, represents the first input torque at the current time, and is a constant greater than 0, is the filtered error, represents a discontinuous sign function; represents a non-singular terminal sliding mode surface vector, and is a constant greater than 0, is in the range of .

[0011] In a second aspect, the application provides a robot driving control device, comprising: an error determination module configured to calculate a tracking error according to a desired trajectory and an actual running trajectory corresponding to each joint of the articulated robot at a current time; a first input torque determination module configured to calculate the tracking error using a sliding mode control algorithm to obtain a first input torque at the current time; a second input torque determination module configured to take the actual input torque at the previous time and the actual running trajectory as inputs of a dynamic neural network, predict an unknown torque in a robot dynamics model at the current time, and take the predicted unknown torque as a second input torque at the current time; a driving module configured to determine an actual input torque at the current time according to the first input torque and the second input torque, and control the articulated robot to move according to the actual input torque at the current time.

[0012] In one or more possible embodiments, the second input torque determination module is specifically configured to: take an input vector Z as an input of the dynamic neural network; wherein the input vector Z includes a system state, the tracking error, and the actual input torque at the previous time; the system state is used to represent the actual running trajectory at the current time; perform nonlinear transformation on the input vector Z through weights and an implicit layer activation function of the dynamic neural network to obtain an approximation output of the unknown torque, and take the approximation output as the second input torque at the current time; wherein the approximation output contains an approximation error, and the approximation error is not greater than a preset value.

[0013] In one or more possible embodiments, further comprising a weight update module configured to update the weights of the dynamic neural network according to a preset update rate, the update rate being related to the tracking error and the actual input torque at the previous time.

[0014] In one or more possible embodiments, the weight update module is specifically configured to: calculate the preset update rate according to the following formula:

[0015] wherein, denotes the preset update rate, is a diagonal matrix gain, and j denotes the number of implicit layer nodes, represents a hidden layer activation function, s is a sliding mode surface vector constructed according to the tracking error, represents a weight of a dynamic neural network, represents a normal constant damping coefficient.

[0016] In one or more possible embodiments, the first input torque determination module is specifically configured to: determine a filtering error according to the tracking error and a derivative of the tracking error; construct a non-singular terminal sliding mode surface based on the filtering error signal, wherein the non-singular terminal sliding mode surface is used to represent a deviation degree between a desired trajectory and an actual running trajectory of the articulated robot; calculate the first input torque at the current time according to the filtering error signal and the non-singular terminal sliding mode surface, using the following formula:

[0017] wherein, represents the first input torque at the current time, and is a constant greater than 0, is the filtering error, represents a discontinuous sign function; represents a non-singular terminal sliding mode surface vector, and is a constant greater than 0, the range of is .

[0018] In a third aspect, the present application provides an electronic device, comprising at least one processor, and a memory in communication connection with the at least one processor, wherein: The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method according to any one of the first aspect.

[0019] In a fourth aspect, the present application provides a computer readable medium storing computer executable instructions for executing the method according to any one of the first aspect.

[0020] According to the robot driving control method, device, equipment and storage medium provided by the present application, the problem of stability and accuracy decline of the articulated robot caused by various uncertain factors in the prior art can be solved. BRIEF DESCRIPTION OF DRAWINGS

[0021] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the application and, together with the description, further serve to explain the principles of the application and, without in any way limiting the present application, illustrate the application.

[0022] Figure 1 A module diagram of a controller design method based on a mechanism model according to an embodiment; Figure 2 A module diagram of a controller design method based on a local data model according to an embodiment; Figure 3 A flowchart of a robot drive control method according to an embodiment; Figure 4 A robot robust control architecture based on data-driven and structure adaptation according to an embodiment; Figure 5 A system architecture diagram according to an embodiment; Figure 6 A curve diagram of a given trajectory, an actual trajectory, and a tracking trajectory error of a link 1 according to an embodiment; Figure 7 A curve diagram of a given trajectory, an actual trajectory, and a tracking trajectory error of a link 2 according to an embodiment; Figure 8 A control input curve diagram of a link 1 and a link 2 according to an embodiment; Figure 9 A corresponding uncertainty factor prediction curve diagram of a link 1 and a link 2 according to an embodiment; Figure 10 A robot drive control device schematic diagram according to an embodiment; Figure 11 A schematic diagram of an electronic device according to an embodiment; Figure 12 A computer readable storage medium schematic diagram according to an embodiment. DETAILED DESCRIPTION

[0023] In order to make the objects, technical solutions and advantages of the present application clearer, the following will further describe the present application with reference to the drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0024] It should be noted that the terms "first", "second", etc. in the description of the present disclosure and claims and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than that illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Rather, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0025] Also, in the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in the text only describes the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two.

[0026] At present, most of the control methods of linear and nonlinear systems are based on models. If the mathematical model of the controlled system cannot be accurately obtained, it will often lead to a decrease in control accuracy, deterioration of dynamic performance, and even instability of the system, making it difficult to achieve the desired control goal.

[0027] The so-called "mechanism" means that the mathematical model is established by theoretical derivation from the physical law, kinematics and dynamics of the system. For example, for mechanical systems, the motion equation can be constructed according to Newton's law or Lagrange equation; for circuit systems, the state equation can be established based on Kirchhoff's law. This modeling method based on mechanism has clear physical meaning, but is often limited in practical application due to system complexity and uncertainty.

[0028] The current controller design mainly follows the following three ideas: one is to design the controller completely based on the mechanism model, the structure and parameters of which are derived from the theoretical model, as shown in Figure 1 ; two is that for systems with inaccurate mechanism model and significant uncertainty, the controller design can introduce data-driven or model-free methods for compensation under the mechanism framework; three is that for systems with complex model, high order, strong nonlinearity, the controller design method based on local data model can be used to reduce the dependence on the global accurate model, as shown in Figure 2The so-called "mechanism" refers to the process of establishing a mathematical model through theoretical derivation from the physical laws, kinematics and dynamics principles of the system. For example, for a mechanical system, its motion equation can be constructed according to Newton's law or Lagrange equation; for a circuit system, its state equation can be established based on Kirchhoff's law. This mechanism-based modeling method has clear physical meaning, but is often limited in practical applications due to system complexity and uncertainty.

[0029] With the development of industry, the production scale expands, the equipment integration improves, and the technical demand becomes increasingly complex, making it very difficult to establish an accurate system mathematical model. Especially for robot systems, the modeling process must face various uncertainties: including structural uncertainty determined by its own properties, which is difficult to accurately estimate in motion; non-structural uncertainty caused by measurement error, parameter perturbation, internal friction, unmodeled high-order dynamics, etc.; and environmental uncertainty from external environment (such as wind speed, temperature and humidity, gravity change, etc.) interference. These factors make it particularly difficult to accurately model robots, which in turn poses a serious challenge to the design of high-precision and high-robustness controllers; and the effectiveness and accuracy of the robot control driving method largely determine the tracking effect of the system, and the existing control driving methods have the following problems: (1) The modeling of robots must have structural uncertainty, non-structural uncertainty, and environmental uncertainty, which makes it impossible to accurately model the robot, resulting in significant steady-state error or even instability in the actual application of the control method based on the accurate model.

[0030] (2) Robots are sensitive to changes in their own state, and the system has high real-time requirements, so there are strict requirements for the real-time performance and engineering implementation of the control algorithm design, but existing high-precision control algorithms often have heavy computational burden, making it difficult to balance real-time performance and control accuracy.

[0031] (3) The robot control system is a complex system, and there are uncertain time-varying, internal friction, parameter perturbation, external environmental disturbance, and gravity field, etc. factors, so it is necessary to design a reasonable controller to ensure the stability of the system.

[0032] (4) The robot has multiple task states, and how to design a stable controller without being affected by structural changes is a problem that needs to be solved.

[0033] (5) Existing parameter adaptive methods perform outstandingly in fast response and real-time control, but have limited adaptability to environmental changes.

[0034] Robots have been widely used in industrial assembly, machine welding, safety and explosion prevention, engineering machinery and other fields due to their unique operation and flexibility. For different task modes, the spatial motion trajectory of the connecting rod robot needs to be planned to achieve accurate positioning or tracking. Therefore, it is particularly important to accurately, quickly and efficiently control the actuator of the connecting rod robot. The spatial trajectory tracking control problem of the connecting rod robot is to track the given expected trajectory by giving the driving torque between each connecting rod, so as to track the given expected trajectory.

[0035] For the convenience of understanding, the following explains the terms involved in the embodiments of the application: Data-driven control refers to a control theory and method in which the controller is designed without relying on the accurate mechanism model of the controlled object, using online or offline I / O data of the controlled system and knowledge obtained through data processing to design the controller, and having convergence, stability and robustness conclusions under certain assumption conditions; the essence is that the unknown part in the model is equivalent to a dynamic linear model, which is then used to approximate the unknown part in the system. The main advantage is that the mechanism model of the system does not need to be accurately known during controller design, only the I / O data of the system is needed; and the system does not need to be trained in advance, and the system can realize adaptive structure control, which is suitable for industrial systems with complex or difficult-to-model models.

[0036] Sliding mode variable structure control is a nonlinear controller with strong robustness and strong fault tolerance capability for various uncertainties of the system. When the model and parameters of the controlled object vary within a certain range, a sliding mode controller can be designed to achieve good control effect, that is, the system has good fault tolerance capability for parameter variation, model uncertainty and external disturbance. At present, eliminating chattering is one of the main directions of sliding mode control research. Time lag, sign function switching, system inertia and other factors are the causes of system chattering, and the switching of the sign function is also the main cause of system singularity. These causes not only increase the energy loss of the system and reduce the service life of the actuator, but also easily excite the unmodeled dynamics of the system, affecting the control quality of the system.

[0037] Finite time stability is a core research direction in control theory, which requires the state variable to converge to zero accurately after a finite time under certain initial conditions, rather than the infinite time convergence in traditional asymptotic stability. Compared with traditional asymptotic stability (only converges when time tends to infinity), finite time stability has faster convergence speed and stronger anti-disturbance performance, and is an important goal of high-precision and high-dynamic control systems.

[0038] Dynamic neural network: a neural network with structural adaptive ability, the structural adaptation of dynamic neural network refers to the ability of the network to automatically adjust its network structure according to the changes of input data or task requirements. This ability makes the network have higher flexibility and efficiency in dealing with complex and variable tasks. Traditional neural networks usually need to manually adjust hyperparameters and maintain fixed network structure during training, while dynamic neural networks can automatically optimize these parameters through real-time learning and feedback mechanism, which has more advantages in dealing with dynamic and changing tasks.

[0039] Structural adaptation: as the complexity of data and environmental uncertainty increases, traditional fixed structure models are difficult to meet diversified needs. Structural adaptation improves flexibility and robustness by dynamically adjusting model architecture. Structural adaptation refers to the ability of a system or model to adjust its structure according to environmental changes or input data. This ability enables the system to dynamically optimize performance and adapt to different scenarios.

[0040] Time axis synchronization strategy: considering the lag of data transmission and control rhythm, simulating the time delay phenomenon of biological neural network, the invention adopts an adaptive way to deal with time delay problem, so as to keep time axis synchronization. By monitoring system performance indicators (such as error and delay) in real time, the control rate is dynamically adjusted to change the network dynamics, achieving the synchronization goal. Time axis synchronization control refers to ensuring the coordination and consistency of different parts of the network or different networks in time, so as to correctly process time series data or realize stable control of dynamic systems.

[0041] With the development of current industry, the production scale, equipment industry and technical demand become very complex, it is difficult to establish an accurate mathematical model of the system, but the input and output (I / O) data of the system can be obtained. Therefore, the present application is based on data driving and uses observable I / O data to construct a direct description of the dynamic characteristics of the system, thereby breaking the dependence on accurate analytical models. The present application provides a driving control method for a robot, which predicts the unknown torque caused by the uncertainty factors of the system model through a dynamic neural network, that is, obtains the mathematical model of the unknown part of the robot system, realizes robust tracking control with online learning and adaptive ability without relying on accurate dynamic model, and the specific flow chart is as shown in Figure 3 The method comprises the following steps: Step 301, calculating the tracking error according to the expected trajectory and the actual running trajectory of each joint of the articulated robot at the current time; In one or more possible embodiments, a preset rotation angle is given for each joint of the articulated robot Wherein, i represents the number of joints, qi represents the given value of the rotation angle of the i th joint; then the actual rotation angle of each joint is measured by a sensor , get the real measurement value; for the convenience of subsequent calculation, define as the reference trajectory of the system, and satisfy , , , as the actual trajectory of the robot, then the tracking error can be defined as formula (1): (1) Wherein, 1, 2, 3,..., i is a positive integer, there are several joints in the system (link robot), the number of i is several.

[0042] Step 302, the sliding mode control algorithm is adopted to calculate the tracking error, and the first input torque at the current time is obtained; In one or more possible embodiments, the robot model is divided into a local mechanism model and a completely unknown mathematical model, the unknown mathematical model includes: internal and external disturbances, unknown gravity, mutual friction of actuator, etc. The unknown mathematical model cannot be accurately measured and quantified, but the local mechanism model can be obtained by Lagrange mechanism modeling method; formula (2) is a rigid body model of link robot: (2) Wherein, is the inertia matrix of the system, is the Coriolis matrix, is the gravity vector, is the friction vector, is the external disturbance in the geodetic coordinate system, is a constant matrix, is the control input torque, is the system state, represents the number of joints / link, that is, the order of the system, are the joint angular displacement, angular velocity and angular acceleration, respectively, which are all system state vectors; The rigid body model of the link robot has the following properties: Property one: the inertia matrix is positive definite and symmetric, and satisfies: Wherein, and are unknown normal numbers.

[0043] Property two: Wherein, ζ represents an arbitrary, non-zero n-dimensional real number column vector; represents the Euclidean norm of vector ζ.

[0044] In one or more possible embodiments, a filter error is calculated according to the tracking error, specifically as follows, and a filter error is determined according to the tracking error and a derivative of the tracking error: (3) wherein, , , is a diagonal matrix symbol, and according to the structure of equation (3), it can be known that and have the same convergence.

[0045] In one or more possible embodiments, a non-singular terminal sliding mode surface is constructed based on the filter error signal; wherein the non-singular terminal sliding mode surface is used to represent the deviation between the expected trajectory and the actual running trajectory of the articulated robot; and a specific formula of the non-singular terminal sliding mode surface is as follows: (4) wherein, and are greater than 0, the filter error signal r is greater than 1, and sign( ) represents a sign function.

[0046] In one or more possible embodiments, equation (2) is transformed to obtain equation (5): (5) wherein, according to the property one of the rigid body model, is positive definite and symmetric and bounded, then is also positive definite and bounded, and then equation (6) can be obtained (6) wherein, is a constant reversible diagonal matrix, is unknown; Equation (6) decomposes into a constant diagonal matrix A and an unknown term Δm, which is mainly based on the physical characteristics of the robot dynamics model and the control design requirements. Since is positive definite, symmetric and bounded, its inverse matrix exists and is bounded. In order to facilitate the design and implementation of the controller, the system model is often decomposed into a known nominal part and an unknown perturbation part: wherein A is a reversible diagonal matrix, representing the nominal inertia after decoupling of each joint; Δm covers model uncertainties and unmodeled dynamics, providing a basis for subsequent design of robust or adaptive control law, so that the controller can effectively compensate for system disturbances and parameter changes while utilizing prior model information.

[0047] In one or more possible embodiments, the formula (4) is calculated and derived to obtain formula (7): (7) The formula (5) is substituted into the formula (7) to obtain formula (8): (8) Further, the formula (6) is substituted into the formula (8) to obtain formula (9): (9) Wherein: , , ; based on physical facts, it is generally reasonable to assume that external disturbances are bounded, so represents the maximum possible value of external disturbance, which is used to represent that the external disturbance is bounded; in the design of the robot control system, the external disturbance is a key factor affecting the tracking accuracy and system stability, and concentrates the influence of the external disturbance and separates it from the system model, so as to facilitate the controller design, represents the external disturbance in the earth coordinate system After the inverse of the system inertia matrix mapping, the equivalent disturbance term (according to formula (5)) generated at the joint acceleration level is obtained; In one or more possible embodiments, formula (9) is transformed to obtain the following formula (10): (10) Wherein, δ represents the input of the control input torque, that is, only the calculated control input torque needs to be input and applied to the system to ensure the stability and tracking accuracy of the closed-loop system; the calculable term in the formula is taken as the first input torque at the current time: , represents the first input torque at the current time, and are constants greater than 0, is the filtering error, represents the discontinuous sign function; represents the non-singular terminal sliding mode surface vector, and are constants greater than 0, the range of ; finally, represents the predicted approximation of the uncertain term, and the specific calculation method is described later.

[0048] Step 303: Use the actual input torque and the actual running trajectory of the previous moment as inputs to the dynamic neural network to predict the unknown torque in the robot dynamics model at the current moment, and use the predicted unknown torque as the second input torque at the current moment. In one or more possible embodiments, the input vector Z is used as the input to a dynamic neural network; wherein the input vector Z includes the system state, the tracking error, and the actual input torque at the previous moment; the system state is used to characterize the actual running trajectory at the current moment; the input vector Z is nonlinearly transformed by the weights and hidden layer activation functions of the dynamic neural network to obtain an approximate output for the unknown torque, and the approximate output is used as the second input torque at the current moment; wherein the approximate output includes an approximate error, and the approximate error is not greater than a preset value; For unknown functions A dynamic neural network can be designed to approximate it; an ideal adaptive dynamic neural network can be designed as shown in formula (11), as follows: (11) in: Neural network weights, where T represents the transpose of the matrix; For the hidden layer activation function, this application uses a Gaussian function. Where is the number of hidden layer nodes, P represents the number of joints / links (i.e., the system order), and Z is the input to the neural network. ,in These represent joint angular displacement, angular velocity, and angular acceleration, respectively, all of which are system state vectors. It is worth noting that vector Z contains... This represents the actual input torque at the previous moment, not the actual input torque at the current moment; To approximate the error, we have: ; In one or more possible embodiments, the actual adaptive dynamic neural network can be designed as follows (12): (12) in: This represents the unknown parts of the predicted model, such as parameter perturbations, unmodeled dynamics, nonlinear coupling terms (Coriolis force, centrifugal force, gravity, friction), and external disturbances. It is the approximate output of the unknown torque, and the approximate output is used as the second input torque at the current moment. Let Z represent the Gaussian function, and let Z be the input to the neural network. To approximate the error; This indicates the preset update rate. The obtained neural network weights, This indicates the preset update rate. is diagonal gain matrix, j denotes the number of hidden layer nodes, denotes the activation function of hidden layer, s is the sliding mode surface vector constructed according to the tracking error, denotes the weight of dynamic neural network, denotes the normal damping coefficient.

[0049] In one or more possible embodiments, the weight of the dynamic neural network is updated according to a preset update rate, the update rate is related to the tracking error and the actual input torque at the previous time; the preset update rate is specifically calculated according to the following formula:

[0050] wherein, denotes the preset update rate, is gain matrix, j denotes the number of hidden layer nodes, denotes the activation function (Gaussian function) of hidden layer, the input of neural network is Z, wherein are respectively joint angle displacement, angular velocity and angular acceleration, all of which are system state vectors; it is worth noting that denotes the actual input torque at the previous time, not the actual input torque at the current time, s is the sliding mode surface vector constructed according to the tracking error, denotes the weight of dynamic neural network, denotes the normal damping coefficient.

[0051] In one or more possible embodiments, in the robot adaptive control design based on dynamic neural network, the "ideal" and "actual" neural network models are respectively given, which is a typical expression method considering both theoretical completeness and engineering realizability, the purpose is to transit from theoretical hypothesis to implementable algorithm, so as to build a robust controller with strict stability guarantee and online learning; first, the ideal neural network model is expressed as ; the model theoretically assumes that there is a set of optimal weights W and enough hidden layer nodes, so that the neural network can approximate the lumped uncertainty N in the system with a bounded error (meeting ). The "ideal" here does not mean that it can be directly implemented, but to provide a reference performance upper bound in subsequent stability analysis, indicating that the uncertainty of the system is essentially approximable by the neural network structure, and the approximation error is controllable, thereby laying a foundation for the stability proof of the entire closed-loop system; secondly, the actual neural network model introduces the online adaptive update law of the weight on the basis of the ideal structure: The design of the above update rate has a clear engineering significance: the first term The weight value is adjusted in real time according to the current sliding mode variable s, so that the neural network can dynamically track the change of system uncertainty, and realize online learning and real-time compensation of unknown dynamics; the second term is a damping term (a correction term), which prevents the weight value from drifting or unbounded growth when the continuous excitation is insufficient or the tracking error is small, thereby enhancing the robustness of the neural network in long-term operation and avoiding over-parameterization or divergence; in summary, the feasibility of neural network approximation and the error boundary are first theoretically established, and then an actual adaptive law with online learning and self-stabilization mechanism is designed, which not only ensures that the control system meets the stability requirements in the Lyapunov sense in theory, but also provides an algorithm structure that can be calculated and executed in real time in the actual system, thereby effectively solving the tracking accuracy and robustness problems caused by model uncertainty, parameter perturbation and external disturbance in robot control.

[0052] In step 304, the actual input torque at the current time is determined according to the first input torque and the second input torque, and the link robot is controlled to move according to the actual input torque at the current time.

[0053] In one or more possible embodiments, the formula calculated according to the first input torque and the second input torque is formula (10), specifically: (10) In formula (10), the first input torque is ; contains two levels of effects: the power term based on the auxiliary variable r (i.e., the filtering error) ensures that the tracking error converges quickly in a limited time; and the linear term and the power term based on the sliding mode variable s jointly constitute a continuous sliding mode control law, which aims to ensure strong robustness while significantly reducing the high-frequency chattering caused by the traditional sign function sign(s), thereby reducing energy loss and prolonging the service life of the actuator; the second input torque is , which represents the real-time approximation value of the system lumped uncertainty N by the dynamic neural network, and through online learning and adaptive adjustment of the weight value, it can predict and compensate the unknown torque caused by model uncertainty, parameter perturbation, nonlinear friction and external disturbance, thereby effectively reducing the model dependence and improving the adaptability of the controller to complex dynamic environments; finally, the nominal model handles the known nominal dynamics; compensates for unknown dynamics online; and O provides robustness and suppresses chattering. By combining model-based compensation, data-driven learning and robust control, this method can achieve high-precision and smooth robust tracking of the desired trajectory while ensuring global stability and limited-time convergence.

[0054] In one or more possible embodiments, such as Figure 4 As shown, this invention proposes a robust robot control architecture based on data-driven and structurally adaptive approaches. First, the controller (corresponding to Equation 10) integrates the desired trajectory and system feedback to calculate the actual input torque, forming control commands. This actual input torque incorporates nominal model compensation, robust sliding mode control, and neural network approximation terms, and acts on the rigid body model (i.e., the robot's dynamics model) to drive the actual system motion. The lumped uncertainty N (containing all unmodeled dynamics and disturbances) is separated and approximated online from the rigid body dynamics in real time, and learned and predicted using a structurally adaptive dynamic neural network, outputting... As an estimate of the unknown part N, the output will be... The data is fed back to the controller in real time to recalculate the actual input torque and generate control commands, forming a complete closed loop of "control-execution-learning-compensation". This architecture replaces the reliance on accurate models with data-driven online learning and combines a time axis synchronization strategy to ensure the real-time performance of predictions, ultimately achieving high-precision and robust tracking control under model uncertainty and external disturbances.

[0055] In one or more possible embodiments, substituting equations (10), (11), and (12) into equation (9) yields a closed-loop system: (13) in: Neural network weights, It is an online approximation of the weights of a neural network. It is the neural network weight approximation error. .

[0056] Therefore, for the known model (1) of the linkage robot, the controller is designed as (10), and the final closed-loop system is (13).

[0057] In one or more possible embodiments, such as Figure 5As shown, according to the control method given in the present application, a complete "physical system-digital model-learning prediction" system architecture diagram is provided, which clearly shows the implementation path from the partially known physical model to the fully data-driven intelligent compensation, clearly divides the complete dynamics model of the robot into "known nonlinear model" and "unknown nonlinear model", and performs online learning and synchronous prediction on the unknown part through the data-driven method; The part of the body has a built-in nonlinear model: it refers to the part that can be accurately described by mechanism modeling such as Lagrange method (partial mechanism model); The part of the body has not built a nonlinear model (internal state / unmodeled dynamics): it refers to all uncertainties that cannot be accurately described by fixed mechanism model due to parameter perturbation, complex friction, assembly error and environmental coupling (such as unstructured terrain contact force); The prediction of the data-driven unknown model in the present application is to predict the unmodeled dynamics, which is determined according to the actual rotation angle of the robot joint and the control instruction (given rotation angle, calculated actual input torque at the last time) and other real-time I / O data; Using real-time I / O data, a structure-adaptive dynamic neural network is used to learn and approximate the unmodeled dynamics online, which can adaptively compensate for the inherent delay of the system caused by sensing, communication and calculation, ensure that the prediction output of the digital model is strictly synchronized with the current state of the physical robot, and thus provide accurate and timely compensation signals for real-time control; The controller combines the known part from the mechanism model with the unknown part estimated from the data-driven predictor to generate the final control torque, which drives each joint to achieve accurate rotation angle tracking; Finally, an intelligent control system that can adapt to uncertainties online and has strong robustness is constructed, thereby effectively solving the precision and stability problems faced by traditional model-dependent control in complex and variable environments.

[0058] To ensure that the drive control method of the present application can realize the stability of the closed-loop system, the following two aspects are used for proof: Proof 1: all signals of the system are bounded.

[0059] The Lyapunov function is selected as: (14) wherein, represents the selected Lyapunov candidate function, which is a scalar function used to analyze the stability of the closed-loop system, and by proving that the function is positive definite and its derivative is negative, the conclusion of system stability can be drawn; s represents the sliding mode surface vector, which is a key indicator for measuring the tracking performance of the system, and s=0 means that the system has reached the ideal tracking state; represents the transpose of s; The neural network weight approximation error is calculated as: , The neural network weight, is the online approximation of the neural network weights; is the gain matrix is the inverse matrix of denotes the trace operation of a matrix, i.e., the sum of the diagonal elements of a matrix; is the time derivative of Substituting equation (13) into equation (14) gives equation (15): (15) If then Therefore, there exists and Thus, according to equation (15), equation (16) is obtained: (16) Therefore, the weight update rate of the adaptive dynamic neural network is: (17) Substituting equation (17) into equation (16) gives equation (18): (18) Since and where is the Frobenius norm of a matrix, which quantifies the size or energy of a matrix; represents a known upper bound of the ideal neural network weight size; therefore, according to equation (18), equation (19) is further obtained: (19) where in equation (19): and are positive constants and are positive definite diagonal matrices, then the minimum eigenvalue of , the minimum eigenvalue of , we get: ; denotes a vector, the i-th component is , i is a positive integer, and the number of joints is equal to the number of i; therefore, , p represents the number of joints; since , we have For the sliding mode surface vector s, since , we have , which gives ; and According to Cauchy-Schwarz inequality and and , we have and where, represents the maximum value of external disturbance; where Y in equation (19) is expressed by equation (19-1): (19-1) If there exists: (20) Further, equation (20) is obtained: (21) As long as one of the conditions in equation (20) is satisfied, we have , so that Thus, it can be concluded that the system is ultimately consistent and bounded, and the bound is: (22) (23) According to proof 1, all signals in system (13) are bounded, so we have: holds; where, represents a known upper bound constant of the output vector of the neural network activation function.

[0060] Proof 2: Based on the fact that all signals in the system are bounded, we achieve the finite-time convergence of the system, i.e., global stability.

[0061] The Lyapunov function is chosen as: (24) Taking the derivative of equation (24) gives: (25) where , is the maximum value of the superposition of disturbance and approximation error.

[0062] Since:

[0063] Therefore, equation (24) can be written as: (26) According to Young's inequality , equation (27) is obtained: (27) Assuming , equation (26) can be written as: (28) Further can be written as: (29) According to formula (29) further derivation, formula (30) is obtained: (30) For formula (30), Indicates the Lyapunov function At the initial time t=0, the value is: , Q is an arbitrary given, small positive number; There is a finite time T, so that when , there is, Constantly, a larger Can be selected, so that the system error tends to zero domain in a finite time.

[0064] In order to verify the beneficial effect of the controller designed in the application, the double link planar robot model based on Matlab / Simulink is used for simulation.

[0065] The specific model parameters are:

[0066] Among them:

[0067] The design given value is: , the amplitude unit is radian (rad).

[0068] The simulation control parameters are: in formula (3), the set constant , in formula (4), the set constant , in formula (10), , The dynamic neural network approximation is designed, so for the open loop system (9), the controller is designed: (10) The uncertainty of the system model The time axis is synchronized with the prediction, and in formula (12), Is the 11th order neural network node, , Is the gain matrix.

[0069] The design given trajectory, actual trajectory, tracking trajectory error of link 1 are as follows:​Figure 6 As shown in the figure, Figure 6 As can be seen from the figure, the given trajectory and the actual trajectory of the connecting rod 1 are basically coincided, the tracking error of the connecting rod robot joint 1 can be up to 10 -3 orders of magnitude (rad), and the system can be stabilized within 1 second; The given trajectory, the actual trajectory and the tracking trajectory error of the connecting rod 2 are specifically as shown in the figure, Figure 7 As shown in the figure, Figure 7 As can be seen from the figure, the given trajectory and the actual trajectory of the connecting rod 2 are basically coincided, the tracking error of the connecting rod robot joint 2 can be up to 10 -3 orders of magnitude (rad), and the system can be stabilized within 1 second; As shown in the figure, Figure 8 As can be seen from the figure, the control input of the controller designed by the application is bounded, which reflects that the input energy is bounded in theory and experiment, that is, the input energy can be predicted; As shown in the figure, Figure 9 The simulation curves between the estimated N and the actual N of the connecting rods 1 and 2 are given, and different colors are used for distinction, and the uncertainty of the system model is predicted by the following formula: The time axis synchronous prediction is performed by , that is, the mathematical model of the unknown part of the system is obtained, and Figure 9 As can be seen from the figure, The estimated N can follow the N synchronously without time delay, and The time axis synchronous prediction realizes automatic optimization of the feedback mechanism by means of the first-order derivative, so as to realize structure self-adaption of the unknown model; According to the robot driving control method provided by the application, the beneficial effects are as follows: (1) The application breaks the severe dependence on the system model when designing the controller, and designs a data-driven control method based on time axis synchronization, and by means of the structure self-adaptive dynamic neural network method (formula (12), (17)), the uncertainty of the system model is time axis synchronously predicted by , and the mathematical model of the unknown part of the system is obtained; (2) By dynamically designing the weight update rate of the neural network, the shortcomings that the traditional neural network usually needs manual adjustment of hyperparameters and keeps a fixed network structure during training are avoided, and the dynamic neural network of the application can automatically optimize these parameters through real-time learning and feedback mechanism, so as to realize structure self-adaption; (3) By dynamically adjusting the weight update rate to change the network dynamics characteristics, the time axis synchronization target is realized by monitoring the error, control input energy and other performance indicators of the system in real time, and as can be seen from formula (17), the input of the Gaussian function is Control input of closed loop system related to system error e The target of the present application is to make e tend to 0, and the control input is bounded, that is, the input energy is controllable, so the present application is based on the index of e, so as to dynamically adjust the weight update rate to change the dynamic characteristics of the neural network, and realize the time axis synchronization target; (4) The stability of the system is improved through the non-singular terminal sliding mode, the sliding mode surface is designed as formula (4), the singularity problem that may occur in the traditional terminal sliding mode is eliminated through the improved sliding mode surface design, while the finite time convergence characteristics are retained, the influence on the stability analysis of the closed loop system caused by directly using the sign function is avoided, and the shortcomings that the traditional sliding mode controller needs to know the internal and external disturbances and the uncertainty boundary are avoided; (5) The controller designed in the present application (formula (10)) can realize closed loop stable control on the closed loop system (formula (13)) obtained from the partial known model (formula (1)) of the link robot, the Lyapunov function (formula (14)) is designed, it is proved that all signals in the closed loop system are bounded, the Lyapunov function (formula (24)) is designed, it is proved that the system is in finite time convergence, that is, global stability. The advantage of finite time convergence to 0 is fast response, high precision control, and enhanced system stability; (6) Input energy is predictable: at present, the controller is designed from the stable execution of the output end, and the controller is not started from the energy of the input end, the controller designed in the present application is based on the input and output data of the system, and the control input It is theoretically proved that all signals in the system are bounded, that is, the input energy is essentially bounded, that is, the input energy is predictable; (7) The nonlinear model of the link robot built based on Matlab / Simulink is adopted to verify the effectiveness of the control algorithm designed in the present application.

[0070] The application also provides a robot driving control device, as shown in Figure 10 comprises: An error determination module 1001 is configured to calculate a tracking error according to the expected trajectory and the actual running trajectory of each joint of the link robot at the current time; A first input torque determination module 1002 is configured to calculate the tracking error by using a sliding mode control algorithm, and obtain the first input torque at the current time; The second input torque determination module 1003 is configured to take the actual input torque at the previous moment and the actual running trajectory as inputs of the dynamic neural network, predict the unknown torque in the robot dynamics model at the current moment, and take the predicted unknown torque as the second input torque at the current moment. The driving module 1004 is configured to determine the actual input torque at the current moment according to the first input torque and the second input torque, and control the motion of the serial robot according to the actual input torque at the current moment.

[0071] In one or more possible embodiments, the second input torque determination module is specifically configured to: take an input vector Z as an input of the dynamic neural network, wherein the input vector Z includes a system state, the tracking error, and the actual input torque at the previous moment, and the system state is used to represent the actual running trajectory at the current moment; perform nonlinear transformation on the input vector Z through weights and an implicit layer activation function of the dynamic neural network to obtain an approximation output of the unknown torque, and take the approximation output as the second input torque at the current moment, wherein the approximation output contains an approximation error, and the approximation error is not greater than a preset value.

[0072] In one or more possible embodiments, the method further includes a weight updating module configured to update the weights of the dynamic neural network according to a preset updating rate, and the updating rate is related to the tracking error and the actual input torque at the previous moment.

[0073] In one or more possible embodiments, the weight updating module is specifically configured to: calculate the preset updating rate according to the following formula:

[0074] wherein, represents the preset updating rate, is a diagonal matrix gain, j represents the number of implicit layer nodes, represents an implicit layer activation function, s is a sliding mode surface vector constructed according to the tracking error, represents the weights of the dynamic neural network, represents a normal number damping coefficient.

[0075] In one or more possible embodiments, the first input torque determination module is specifically configured to: determine a filtering error according to the tracking error and a derivative of the tracking error. construct a non-singular terminal sliding mode surface based on the filtered error signal; wherein the non-singular terminal sliding mode surface is used to represent a degree of deviation between a desired trajectory and an actual running trajectory of the articulated robot; According to the filtered error signal and the non-singular terminal sliding mode surface, a first input torque at the current time is calculated by using the following formula:

[0076] wherein, represents the first input torque at the current time, and is a constant greater than 0, is a filtered error, represents a discontinuous sign function; represents a non-singular terminal sliding mode surface vector, and is a constant greater than 0, the range of is .

[0077] The application also provides an electronic device, comprising at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the robot driving control method described above.

[0078] As shown in Figure 11 , the device comprises a processor 1101, a memory 1102, a communication interface 1103 and a bus 1104. Among them, the processor 1101, the memory 1102 and the communication interface 1103 are connected with each other through the bus 1104.

[0079] The processor 1101 is used to read the instructions in the memory 1102 and execute them, so that the at least one processor can perform the robot driving control method provided by the above-mentioned embodiments.

[0080] The memory 1102 is used to store various instructions and programs of the robot driving control method provided by the above-mentioned embodiments.

[0081] The bus 1104 can be a peripheral component interconnect (peripheral component interconnect, abbreviated as PCI) bus or an extended industry standard architecture (extended industry standard architecture, abbreviated as EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For the convenience of representation, Figure 11 only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0082] The processor 1101 can be a central processing unit (CPU), a network processor (NP), a Graphic Processing Unit (GPU), or any combination of CPU, NP, GPU. It can also be a hardware chip. The hardware chip can be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0083] In addition, the present application also provides a computer readable storage medium, such as a computer storage medium, a computer program product, a computer readable non-volatile storage medium, or a computer readable storage medium. Figure 12 As shown, the computer storage medium stores a computer program, and the computer program is used to make the computer execute any one of the above-mentioned methods.

[0084] The memory can include a readable medium in the form of a volatile memory, such as a random access memory (RAM) 1201 and / or a cache memory 1202, and can further include a read-only memory (ROM) 1203.

[0085] The memory can further include a program / utility 1205 having a set of programs / modules 1204, including but not limited to an operating system, one or more applications, other program modules, and program data, each of which or a combination thereof can include implementation of a network environment.

[0086] Those skilled in the art will appreciate that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROMs, optical storage, etc.) containing computer usable program code.

[0087] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0088] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0089] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0090] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A robot drive control method characterized by, The method comprises the following steps: calculating a tracking error according to a desired trajectory and an actual running trajectory corresponding to each joint of the articulated robot at a current time; calculating the tracking error by using a sliding mode control algorithm to obtain a first input torque at the current time; using the actual input torque at the previous time and the actual running trajectory as inputs of a dynamic neural network to predict an unknown torque in a robot dynamics model at the current time, and using the predicted unknown torque as a second input torque at the current time; determining an actual input torque at the current time according to the first input torque and the second input torque, and controlling the articulated robot to move according to the actual input torque at the current time.

2. The method of claim 1, wherein, The method of using the actual input torque at the previous time and the actual running trajectory as inputs of the dynamic neural network to predict the unknown torque in the robot dynamics model at the current time, and using the predicted unknown torque as the second input torque at the current time comprises the following steps: using an input vector Z as inputs of the dynamic neural network; wherein the input vector Z comprises a system state, the tracking error and the actual input torque at the previous time; the system state is used to represent the actual running trajectory at the current time; performing nonlinear transformation on the input vector Z by using weights and hidden layer activation functions of the dynamic neural network to obtain an approximation output of the unknown torque, and using the approximation output as the second input torque at the current time; wherein the approximation output contains an approximation error, and the approximation error is not greater than a preset value.

3. The method of claim 2, wherein, The method further comprises the following steps: updating the weights of the dynamic neural network according to a preset update rate, wherein the update rate is related to the tracking error and the actual input torque at the previous time.

4. The method of claim 3, wherein, The preset update rate is calculated according to the following formula: wherein, denotes a preset update rate, is a diagonal matrix gain, j denotes the number of hidden layer nodes, denotes a hidden layer activation function, s is a sliding mode surface vector constructed according to the tracking error, denotes the weight of the dynamic neural network, denotes a normal number damping coefficient.

5. The method of claim 1, wherein, The method of calculating the tracking error by using the sliding mode control algorithm to obtain the first input torque at the current time comprises the following steps: determining a filtering error according to the tracking error and a derivative of the tracking error; constructing a non-singular terminal sliding mode surface based on the filtering error signal; wherein the non-singular terminal sliding mode surface is used to represent a deviation between a desired trajectory and an actual running trajectory of the articulated robot; calculating the first input torque at the current time according to the filtering error signal and the non-singular terminal sliding mode surface by using the following formula: wherein, represents a first input torque at a current time, and is a constant greater than 0, is a filtered error, represents a discontinuous sign function; represents a non-singular terminal sliding mode surface vector, and is a constant greater than 0, ranges from .

6. A robot drive control device characterized by comprising: The method comprises the following steps: an error determination module is configured to calculate a tracking error according to a desired trajectory and an actual running trajectory corresponding to each joint of the articulated robot at a current time; a first input torque determination module is configured to calculate the tracking error by using a sliding mode control algorithm to obtain a first input torque at the current time; a second input torque determination module is configured to use the actual input torque at the previous time and the actual running trajectory as inputs of a dynamic neural network to predict an unknown torque in a robot dynamics model at the current time, and use the predicted unknown torque as a second input torque at the current time; a driving module is configured to determine an actual input torque at the current time according to the first input torque and the second input torque, and control the articulated robot to move according to the actual input torque at the current time.

7. The apparatus of claim 6, wherein, The second input torque determination module is specifically configured to: input a vector Z as an input of the dynamic neural network, wherein the vector Z comprises a system state, the tracking error, and an actual input torque at a previous time; the system state is used to represent an actual running track at the current time; perform nonlinear transformation on the vector Z through a weight and a hidden layer activation function of the dynamic neural network to obtain an approximation output of the unknown torque, and take the approximation output as the second input torque at the current time; wherein the approximation output comprises an approximation error, and the approximation error is not greater than a preset value.

8. The apparatus of claim 7, wherein, The weight updating module is further configured to update the weight of the dynamic neural network according to a preset updating rate, wherein the updating rate is related to the tracking error and the actual input torque at the previous time.

9. An electronic device, comprising: The method comprises: at least one processor, and a memory connected with the at least one processor in communication, wherein: the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-5.

10. A computer readable medium storing computer executable instructions, wherein the instructions comprise: The computer executable instructions are used to perform the method according to any one of claims 1-5.

Images